sensemaking 0.18.3 → 0.19.0

package/dist/esm/features/embed.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/embed.ts"],"sourcesContent":["import type { Block, Chunk } from '../chunk/index.ts';\nimport { CHUNK_VERSION, chunkFromBlocks, parse } from '../chunk/index.ts';\nimport { embedConfig } from '../config/index.ts';\nimport { modelIdentity } from '../embed/identity.ts';\nimport type { Feature } from './types.ts';\n\n// int8 vectors with a per-vector scale, NULL vector = not yet embedded. Reconcile writes dirty\n// rows; embedding tops up on the next search, so staleness costs recall. Providers, the vector\n// fill, and the search-time helpers live under src/embed/. Block/group structure (D1-D5) lives\n// in src/chunk; this keeps only the title/summary prefix and the empty-chunk drop (D6).\n\nexport type { Chunk };\n\n// Deterministic, so embed time can re-derive text from stored line ranges; the title/summary\n// prefix mirrors bm25 field weighting, and an empty chunk is dropped before the prefix is added.\n//\n// Chunks the body, not the raw file, with `offset` shifting line numbers back onto the raw\n// file so a range stays a direct Read range (sections is 1-indexed over raw too). Chunking raw\n// put the frontmatter block in its own leading chunk on every note with a heading, which made\n// `lines` point at YAML and made frontmatter-only notes near-identical to each other. It also\n// disagreed with FTS, which indexes the body alone.\nfunction chunksOf(blocks: Block[], body: string, search?: { title: string; summary: string }, offset = 0, chunkTokens?: number): Chunk[] {\n  const prefix = [search?.title, search?.summary].filter(Boolean).join('\\n');\n  // chunk()'s default text mode is 'raw' (D9), the shipped choice. A hardcoded option change\n  // here must bump src/chunk/version.ts's CHUNK_VERSION -- the signature for every chunker\n  // input not already carried in config.\n  return chunkFromBlocks(blocks, body, chunkTokens !== undefined ? { targetTokens: chunkTokens } : undefined).map((c) => ({\n    startLine: c.startLine + offset,\n    endLine: c.endLine + offset,\n    text: prefix ? `${prefix}\\n${c.text}` : c.text,\n  }));\n}\n\nexport const embed: Feature = {\n  name: 'embed',\n  async schema(db) {\n    await db.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector BLOB, PRIMARY KEY (\"path\", chunk))`);\n  },\n  extract(raw, body, search, _data, cfg, blocks) {\n    // Lines the frontmatter occupies, so body line 1 maps back to its raw line number.\n    // blocks comes from parseFile's shared parse; falls back to parsing body for a direct call.\n    return chunksOf(blocks ?? parse(body), body, search, raw.split('\\n').length - body.split('\\n').length, cfg?.embed?.chunkTokens);\n  },\n  async remove(db, paths) {\n    if (paths.length === 0) return;\n    await db.runBatch(\n      'DELETE FROM embeddings WHERE \"path\" = ?',\n      paths.map((p) => [p])\n    );\n  },\n  // A tree with no embedding model never had extract() run for the doc (reconcile.ts's per-file\n  // filter skips it), so extracted is undefined here -- those docs contribute no rows.\n  async store(db, docs) {\n    const rows: unknown[][] = [];\n    for (const { path, extracted } of docs) {\n      if (!extracted) continue;\n      (extracted as Chunk[]).forEach((c, idx) => rows.push([path, idx, c.startLine, c.endLine]));\n    }\n    if (rows.length === 0) return;\n    await db.runBatch('INSERT INTO embeddings (\"path\", chunk, start_line, end_line, scale, vector) VALUES (?, ?, ?, ?, NULL, NULL)', rows);\n  },\n  enabledForFile(_cfg, file) {\n    return file.embed;\n  },\n  // Provider/model/chunk-version/identity, keyed so a rebuild notice can name it \"embed\n  // settings\". Driven by embedConfig(cfg) directly (a model can be configured with no preset\n  // yet using vectors), not by whether this feature is itself active.\n  signature(cfg) {\n    const e = embedConfig(cfg);\n    if (!e) return 'embed:off';\n    // Weight identity from identity.ts, no network: a static model's resolved sha or local\n    // size+mtime, appended once known so changed weights re-embed.\n    const identity = e.provider === 'static' ? modelIdentity(e.model) : undefined;\n    // W3b: chunkTokens rides the version token, since it's the one owner lever the chunker\n    // itself takes -- changing or clearing it must rebuild exactly like a chunker version bump.\n    const chunkVersion = e.chunkTokens !== undefined ? `${CHUNK_VERSION}:${e.chunkTokens}` : CHUNK_VERSION;\n    return `embed:${e.provider}:${e.model}:${chunkVersion}${identity !== undefined ? `@${identity}` : ''}`;\n  },\n};\n"],"names":["CHUNK_VERSION","chunkFromBlocks","parse","embedConfig","modelIdentity","chunksOf","blocks","body","search","offset","chunkTokens","prefix","title","summary","filter","Boolean","join","undefined","targetTokens","map","c","startLine","endLine","text","embed","name","schema","db","exec","extract","raw","_data","cfg","split","length","remove","paths","runBatch","p","store","docs","rows","path","extracted","forEach","idx","push","enabledForFile","_cfg","file","signature","e","identity","provider","model","chunkVersion"],"mappings":"AACA,SAASA,aAAa,EAAEC,eAAe,EAAEC,KAAK,QAAQ,oBAAoB;AAC1E,SAASC,WAAW,QAAQ,qBAAqB;AACjD,SAASC,aAAa,QAAQ,uBAAuB;AAUrD,6FAA6F;AAC7F,iGAAiG;AACjG,EAAE;AACF,2FAA2F;AAC3F,+FAA+F;AAC/F,8FAA8F;AAC9F,8FAA8F;AAC9F,oDAAoD;AACpD,SAASC,SAASC,MAAe,EAAEC,IAAY,EAAEC,MAA2C,EAAEC,SAAS,CAAC,EAAEC,WAAoB;IAC5H,MAAMC,SAAS;QAACH,mBAAAA,6BAAAA,OAAQI,KAAK;QAAEJ,mBAAAA,6BAAAA,OAAQK,OAAO;KAAC,CAACC,MAAM,CAACC,SAASC,IAAI,CAAC;IACrE,2FAA2F;IAC3F,yFAAyF;IACzF,uCAAuC;IACvC,OAAOf,gBAAgBK,QAAQC,MAAMG,gBAAgBO,YAAY;QAAEC,cAAcR;IAAY,IAAIO,WAAWE,GAAG,CAAC,CAACC,IAAO,CAAA;YACtHC,WAAWD,EAAEC,SAAS,GAAGZ;YACzBa,SAASF,EAAEE,OAAO,GAAGb;YACrBc,MAAMZ,SAAS,GAAGA,OAAO,EAAE,EAAES,EAAEG,IAAI,EAAE,GAAGH,EAAEG,IAAI;QAChD,CAAA;AACF;AAEA,OAAO,MAAMC,QAAiB;IAC5BC,MAAM;IACN,MAAMC,QAAOC,EAAE;QACb,MAAMA,GAAGC,IAAI,CAAC,CAAC,8JAA8J,CAAC;IAChL;IACAC,SAAQC,GAAG,EAAEvB,IAAI,EAAEC,MAAM,EAAEuB,KAAK,EAAEC,GAAG,EAAE1B,MAAM;YAG4D0B;QAFvG,mFAAmF;QACnF,4FAA4F;QAC5F,OAAO3B,SAASC,mBAAAA,oBAAAA,SAAUJ,MAAMK,OAAOA,MAAMC,QAAQsB,IAAIG,KAAK,CAAC,MAAMC,MAAM,GAAG3B,KAAK0B,KAAK,CAAC,MAAMC,MAAM,EAAEF,gBAAAA,2BAAAA,aAAAA,IAAKR,KAAK,cAAVQ,iCAAAA,WAAYtB,WAAW;IAChI;IACA,MAAMyB,QAAOR,EAAE,EAAES,KAAK;QACpB,IAAIA,MAAMF,MAAM,KAAK,GAAG;QACxB,MAAMP,GAAGU,QAAQ,CACf,2CACAD,MAAMjB,GAAG,CAAC,CAACmB,IAAM;gBAACA;aAAE;IAExB;IACA,8FAA8F;IAC9F,qFAAqF;IACrF,MAAMC,OAAMZ,EAAE,EAAEa,IAAI;QAClB,MAAMC,OAAoB,EAAE;QAC5B,KAAK,MAAM,EAAEC,IAAI,EAAEC,SAAS,EAAE,IAAIH,KAAM;YACtC,IAAI,CAACG,WAAW;YACfA,UAAsBC,OAAO,CAAC,CAACxB,GAAGyB,MAAQJ,KAAKK,IAAI,CAAC;oBAACJ;oBAAMG;oBAAKzB,EAAEC,SAAS;oBAAED,EAAEE,OAAO;iBAAC;QAC1F;QACA,IAAImB,KAAKP,MAAM,KAAK,GAAG;QACvB,MAAMP,GAAGU,QAAQ,CAAC,+GAA+GI;IACnI;IACAM,gBAAeC,IAAI,EAAEC,IAAI;QACvB,OAAOA,KAAKzB,KAAK;IACnB;IACA,sFAAsF;IACtF,2FAA2F;IAC3F,oEAAoE;IACpE0B,WAAUlB,GAAG;QACX,MAAMmB,IAAIhD,YAAY6B;QACtB,IAAI,CAACmB,GAAG,OAAO;QACf,uFAAuF;QACvF,+DAA+D;QAC/D,MAAMC,WAAWD,EAAEE,QAAQ,KAAK,WAAWjD,cAAc+C,EAAEG,KAAK,IAAIrC;QACpE,uFAAuF;QACvF,4FAA4F;QAC5F,MAAMsC,eAAeJ,EAAEzC,WAAW,KAAKO,YAAY,GAAGjB,cAAc,CAAC,EAAEmD,EAAEzC,WAAW,EAAE,GAAGV;QACzF,OAAO,CAAC,MAAM,EAAEmD,EAAEE,QAAQ,CAAC,CAAC,EAAEF,EAAEG,KAAK,CAAC,CAAC,EAAEC,eAAeH,aAAanC,YAAY,CAAC,CAAC,EAAEmC,UAAU,GAAG,IAAI;IACxG;AACF,EAAE"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/embed.ts"],"sourcesContent":["import type { Block, Chunk } from '../chunk/index.ts';\nimport { CHUNK_VERSION, chunkFromBlocks, parse } from '../chunk/index.ts';\nimport { embedConfig } from '../config/index.ts';\nimport { stashChunkText } from '../embed/handoff.ts';\nimport { modelIdentity } from '../embed/identity.ts';\nimport { countLines } from '../scan/frontmatter.ts';\nimport type { Feature } from './types.ts';\n\n// int8 vectors with a per-vector scale, NULL vector = not yet embedded: reconcile writes dirty\n// rows and embedding tops up on the next search, so staleness costs recall.\n\nexport type { Chunk };\n\n// The title/summary prefix mirrors bm25 field weighting. Chunks the body, not the raw file, so\n// `offset` shifts line numbers back onto raw and a range stays a direct Read range.\nfunction chunksOf(blocks: Block[], body: string, search?: { title: string; summary: string }, offset = 0, chunkTokens?: number): Chunk[] {\n  const prefix = [search?.title, search?.summary].filter(Boolean).join('\\n');\n  // A hardcoded option change here must bump src/chunk/version.ts's CHUNK_VERSION, the signature\n  // for every chunker input not already carried in config.\n  return chunkFromBlocks(blocks, body, chunkTokens !== undefined ? { targetTokens: chunkTokens } : undefined).map((c) => ({\n    startLine: c.startLine + offset,\n    endLine: c.endLine + offset,\n    text: prefix ? `${prefix}\\n${c.text}` : c.text,\n  }));\n}\n\nexport const embed: Feature = {\n  name: 'embed',\n  async schema(db) {\n    await db.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector BLOB, PRIMARY KEY (\"path\", chunk))`);\n  },\n  extract(raw, body, search, _data, cfg, blocks) {\n    // Lines the frontmatter occupies, so body line 1 maps back to its raw line number.\n    // blocks comes from parseFile's shared parse; falls back to parsing body for a direct call.\n    return chunksOf(blocks ?? parse(body), body, search, countLines(raw) - countLines(body), cfg?.embed?.chunkTokens);\n  },\n  async remove(db, paths) {\n    if (paths.length === 0) return;\n    await db.runBatch(\n      'DELETE FROM embeddings WHERE \"path\" = ?',\n      paths.map((p) => [p])\n    );\n  },\n  // A tree with no embedding model never had extract() run for the doc (reconcile.ts's per-file\n  // filter skips it), so extracted is undefined here -- those docs contribute no rows.\n  async store(db, docs) {\n    const rows: unknown[][] = [];\n    const texts = new Map<string, string[]>();\n    for (const { path, extracted } of docs) {\n      if (!extracted) continue;\n      const chunks = extracted as Chunk[];\n      chunks.forEach((c, idx) => rows.push([path, idx, c.startLine, c.endLine]));\n      texts.set(\n        path,\n        chunks.map((c) => c.text)\n      );\n    }\n    if (rows.length === 0) return;\n    await db.runBatch('INSERT INTO embeddings (\"path\", chunk, start_line, end_line, scale, vector) VALUES (?, ?, ?, ?, NULL, NULL)', rows);\n    // The text extract() just computed, handed to the embed pass that follows in this same\n    // process rather than re-derived from the file there (src/embed/handoff.ts).\n    stashChunkText(db, texts);\n  },\n  enabledForFile(_cfg, file) {\n    return file.embed;\n  },\n  // Keyed so a rebuild notice can name it \"embed settings\". Driven by embedConfig(cfg) directly:\n  // a model can be configured with no preset yet using vectors.\n  signature(cfg) {\n    const e = embedConfig(cfg);\n    if (!e) return 'embed:off';\n    // Weight identity from identity.ts, no network: a static model's resolved sha or local\n    // size+mtime, appended once known so changed weights re-embed.\n    const identity = e.provider === 'static' ? modelIdentity(e.model) : undefined;\n    // W3b: chunkTokens rides the version token, since it's the one owner lever the chunker\n    // itself takes -- changing or clearing it must rebuild exactly like a chunker version bump.\n    const chunkVersion = e.chunkTokens !== undefined ? `${CHUNK_VERSION}:${e.chunkTokens}` : CHUNK_VERSION;\n    return `embed:${e.provider}:${e.model}:${chunkVersion}${identity !== undefined ? `@${identity}` : ''}`;\n  },\n};\n"],"names":["CHUNK_VERSION","chunkFromBlocks","parse","embedConfig","stashChunkText","modelIdentity","countLines","chunksOf","blocks","body","search","offset","chunkTokens","prefix","title","summary","filter","Boolean","join","undefined","targetTokens","map","c","startLine","endLine","text","embed","name","schema","db","exec","extract","raw","_data","cfg","remove","paths","length","runBatch","p","store","docs","rows","texts","Map","path","extracted","chunks","forEach","idx","push","set","enabledForFile","_cfg","file","signature","e","identity","provider","model","chunkVersion"],"mappings":"AACA,SAASA,aAAa,EAAEC,eAAe,EAAEC,KAAK,QAAQ,oBAAoB;AAC1E,SAASC,WAAW,QAAQ,qBAAqB;AACjD,SAASC,cAAc,QAAQ,sBAAsB;AACrD,SAASC,aAAa,QAAQ,uBAAuB;AACrD,SAASC,UAAU,QAAQ,yBAAyB;AAQpD,+FAA+F;AAC/F,oFAAoF;AACpF,SAASC,SAASC,MAAe,EAAEC,IAAY,EAAEC,MAA2C,EAAEC,SAAS,CAAC,EAAEC,WAAoB;IAC5H,MAAMC,SAAS;QAACH,mBAAAA,6BAAAA,OAAQI,KAAK;QAAEJ,mBAAAA,6BAAAA,OAAQK,OAAO;KAAC,CAACC,MAAM,CAACC,SAASC,IAAI,CAAC;IACrE,+FAA+F;IAC/F,yDAAyD;IACzD,OAAOjB,gBAAgBO,QAAQC,MAAMG,gBAAgBO,YAAY;QAAEC,cAAcR;IAAY,IAAIO,WAAWE,GAAG,CAAC,CAACC,IAAO,CAAA;YACtHC,WAAWD,EAAEC,SAAS,GAAGZ;YACzBa,SAASF,EAAEE,OAAO,GAAGb;YACrBc,MAAMZ,SAAS,GAAGA,OAAO,EAAE,EAAES,EAAEG,IAAI,EAAE,GAAGH,EAAEG,IAAI;QAChD,CAAA;AACF;AAEA,OAAO,MAAMC,QAAiB;IAC5BC,MAAM;IACN,MAAMC,QAAOC,EAAE;QACb,MAAMA,GAAGC,IAAI,CAAC,CAAC,8JAA8J,CAAC;IAChL;IACAC,SAAQC,GAAG,EAAEvB,IAAI,EAAEC,MAAM,EAAEuB,KAAK,EAAEC,GAAG,EAAE1B,MAAM;YAG8C0B;QAFzF,mFAAmF;QACnF,4FAA4F;QAC5F,OAAO3B,SAASC,mBAAAA,oBAAAA,SAAUN,MAAMO,OAAOA,MAAMC,QAAQJ,WAAW0B,OAAO1B,WAAWG,OAAOyB,gBAAAA,2BAAAA,aAAAA,IAAKR,KAAK,cAAVQ,iCAAAA,WAAYtB,WAAW;IAClH;IACA,MAAMuB,QAAON,EAAE,EAAEO,KAAK;QACpB,IAAIA,MAAMC,MAAM,KAAK,GAAG;QACxB,MAAMR,GAAGS,QAAQ,CACf,2CACAF,MAAMf,GAAG,CAAC,CAACkB,IAAM;gBAACA;aAAE;IAExB;IACA,8FAA8F;IAC9F,qFAAqF;IACrF,MAAMC,OAAMX,EAAE,EAAEY,IAAI;QAClB,MAAMC,OAAoB,EAAE;QAC5B,MAAMC,QAAQ,IAAIC;QAClB,KAAK,MAAM,EAAEC,IAAI,EAAEC,SAAS,EAAE,IAAIL,KAAM;YACtC,IAAI,CAACK,WAAW;YAChB,MAAMC,SAASD;YACfC,OAAOC,OAAO,CAAC,CAAC1B,GAAG2B,MAAQP,KAAKQ,IAAI,CAAC;oBAACL;oBAAMI;oBAAK3B,EAAEC,SAAS;oBAAED,EAAEE,OAAO;iBAAC;YACxEmB,MAAMQ,GAAG,CACPN,MACAE,OAAO1B,GAAG,CAAC,CAACC,IAAMA,EAAEG,IAAI;QAE5B;QACA,IAAIiB,KAAKL,MAAM,KAAK,GAAG;QACvB,MAAMR,GAAGS,QAAQ,CAAC,+GAA+GI;QACjI,uFAAuF;QACvF,6EAA6E;QAC7EtC,eAAeyB,IAAIc;IACrB;IACAS,gBAAeC,IAAI,EAAEC,IAAI;QACvB,OAAOA,KAAK5B,KAAK;IACnB;IACA,+FAA+F;IAC/F,8DAA8D;IAC9D6B,WAAUrB,GAAG;QACX,MAAMsB,IAAIrD,YAAY+B;QACtB,IAAI,CAACsB,GAAG,OAAO;QACf,uFAAuF;QACvF,+DAA+D;QAC/D,MAAMC,WAAWD,EAAEE,QAAQ,KAAK,WAAWrD,cAAcmD,EAAEG,KAAK,IAAIxC;QACpE,uFAAuF;QACvF,4FAA4F;QAC5F,MAAMyC,eAAeJ,EAAE5C,WAAW,KAAKO,YAAY,GAAGnB,cAAc,CAAC,EAAEwD,EAAE5C,WAAW,EAAE,GAAGZ;QACzF,OAAO,CAAC,MAAM,EAAEwD,EAAEE,QAAQ,CAAC,CAAC,EAAEF,EAAEG,KAAK,CAAC,CAAC,EAAEC,eAAeH,aAAatC,YAAY,CAAC,CAAC,EAAEsC,UAAU,GAAG,IAAI;IACxG;AACF,EAAE"}

package/dist/esm/features/fences.js

@@ -1,8 +1,5 @@
-// Shared line-fence tracker for tags/sections/embed. Opener: >=3 backticks or tildes at column
-// 0 -- no indent allowance, a deliberate divergence from CommonMark's up-to-3-space rule (see
-// test/unit/features/fences.test.ts DIVERGENCES table). A backtick opener's info string must not itself
-// contain a backtick (spec rule). A closer is a run of the SAME character, length >= the
-// opener's, holding nothing but trailing spaces after the run.
+// Shared line-fence tracker for tags/sections/embed. Opener: >=3 backticks or tildes at column 0, no indent
+// allowance (diverges from CommonMark's up-to-3-space rule). Closer: same char, length >= opener's, trailing spaces only.
 const BACKTICK_OPEN = /^(`{3,})(.*)$/;
 const TILDE_OPEN = /^(~{3,})(.*)$/;
 export function fenceTracker() {
@@ -80,10 +77,8 @@ export function commentTracker() {
         }
     };
 }
-// CommonMark's HTML-block type-6 list (fixed by the spec, not a drifting enumeration): a line
-// starting with an open or close tag of one of these, at column 0 (leading whitespace/`>`
-// blockquote markers tolerated), opens a block that swallows following lines -- tags, links,
-// and comments alike -- until a blank line closes it.
+// CommonMark's HTML-block type-6 tag list (fixed by the spec). A line opening or closing one of these at
+// column 0 (leading whitespace/`>` tolerated) swallows following lines -- tags, links, comments alike -- until a blank line.
 const HTML_BLOCK_TAGS = new Set([
     'address',
     'article',
@@ -160,12 +155,8 @@ const HTML_PRE_CLOSE_RE = /<\/(?:script|pre|style|textarea)>/i; // any of the fo
 // An opening or closing tag at column 0, tag name captured for the lookups above.
 const HTML_BLOCK_OPEN_RE = /^<\/?([a-zA-Z][a-zA-Z0-9]*)(?:[ \t]|\/?>|$)/;
 const BLANK_LINE_RE = /^[ \t>]*$/; // whitespace/blockquote-markers only -- a paragraph break, and what ends a type-6 block
-// Body with fenced-code, HTML-block, and <!-- --> comment regions replaced by spaces (newlines
-// preserved, so line numbers and offsets survive) plus inline code spans masked. The single
-// region masker for tags/links: fence and HTML-block delimiter+content lines are opaque, an
-// unclosed comment dies at its containing block's end (or the next blank line at paragraph
-// level -- never silently to end of file), and a closed comment masks exactly its span.
-// Bodies without a backtick/tilde/`<` fast-path out untouched.
+// Body with fenced-code, HTML-block, and <!-- --> comment regions replaced by spaces (newlines preserved, so line numbers and offsets survive), inline code spans masked too.
+// An unclosed comment dies at its block's end or the next paragraph blank line, never silently to end of file; bodies without a backtick/tilde/`<` fast-path out untouched.
 export function maskRegions(body) {
     if (!/[`~<]/.test(body)) return body;
     const fence = fenceTracker();
@@ -195,7 +186,7 @@ export function maskRegions(body) {
         }
         const uncommented = comment.inComment || line.includes('<!--') ? comment.mask(line) : line;
         if (comment.inComment && BLANK_LINE_RE.test(line)) comment.close(); // paragraph-level: dies at the next blank line
-        // Indented or blockquoted HTML blocks still swallow their content in Obsidian, so the
+        // Indented or blockquoted HTML blocks swallow their content in Obsidian too, so the
         // opener test runs after stripping leading whitespace and > markers.
         const stripped = uncommented.replace(/^[ \t>]*/, '');
         if (stripped[0] === '<') {
@@ -222,9 +213,8 @@ export function maskRegions(body) {
         return uncommented.includes('`') ? maskCodeSpans(uncommented) : uncommented;
     }).join('\n');
 }
-// A code span opens on a run of N backticks and closes at the next run of exactly N -- a
-// shorter or longer run in between is literal text, not a delimiter (CommonMark code spans).
-// Masked with spaces so column positions and tag-boundary whitespace are unaffected.
+// A code span opens on a run of N backticks and closes at the next run of exactly N (CommonMark code spans);
+// a shorter/longer run in between is literal text. Masked with spaces so column positions stay unaffected.
 export function maskCodeSpans(line) {
     let out = '';
     let i = 0;

package/dist/esm/features/fences.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/fences.ts"],"sourcesContent":["// Shared line-fence tracker for tags/sections/embed. Opener: >=3 backticks or tildes at column\n// 0 -- no indent allowance, a deliberate divergence from CommonMark's up-to-3-space rule (see\n// test/unit/features/fences.test.ts DIVERGENCES table). A backtick opener's info string must not itself\n// contain a backtick (spec rule). A closer is a run of the SAME character, length >= the\n// opener's, holding nothing but trailing spaces after the run.\n\nexport interface FenceTracker {\n  feed(line: string): boolean; // true iff this line is a fence delimiter (open or close)\n  readonly inFence: boolean;\n}\n\nconst BACKTICK_OPEN = /^(`{3,})(.*)$/;\nconst TILDE_OPEN = /^(~{3,})(.*)$/;\n\nexport function fenceTracker(): FenceTracker {\n  let inFence = false;\n  let fenceChar = '';\n  let fenceLen = 0;\n  return {\n    feed(line: string): boolean {\n      if (!inFence) {\n        const bt = BACKTICK_OPEN.exec(line);\n        if (bt && !bt[2].includes('`')) {\n          inFence = true;\n          fenceChar = '`';\n          fenceLen = bt[1].length;\n          return true;\n        }\n        const td = TILDE_OPEN.exec(line);\n        if (td) {\n          inFence = true;\n          fenceChar = '~';\n          fenceLen = td[1].length;\n          return true;\n        }\n        return false;\n      }\n      const run = fenceChar === '`' ? /^(`+)(.*)$/ : /^(~+)(.*)$/;\n      const m = run.exec(line);\n      if (m && m[1].length >= fenceLen && m[2].trim() === '') {\n        inFence = false;\n        return true;\n      }\n      return false;\n    },\n    get inFence() {\n      return inFence;\n    },\n  };\n}\n\n// Masks <!-- ... --> spans that may cross line boundaries, one line at a time so callers can\n// interleave it with their own per-line state. In isolation an unclosed <!-- stays open forever\n// (mask() has no notion of a line boundary that should kill it); maskRegions() below is what\n// calls close() at the point an unclosed comment actually dies -- a blank line at paragraph\n// level, or its enclosing HTML block's end.\nexport interface CommentTracker {\n  mask(line: string): string;\n  close(): void; // force-ends an unclosed comment (paragraph blank line, or its HTML block ending)\n  readonly inComment: boolean;\n}\n\nexport function commentTracker(): CommentTracker {\n  let inComment = false;\n  return {\n    mask(line: string): string {\n      let out = '';\n      let i = 0;\n      while (i < line.length) {\n        if (inComment) {\n          const end = line.indexOf('-->', i);\n          if (end === -1) {\n            out += ' '.repeat(line.length - i);\n            i = line.length;\n          } else {\n            out += ' '.repeat(end + 3 - i);\n            i = end + 3;\n            inComment = false;\n          }\n          continue;\n        }\n        const start = line.indexOf('<!--', i);\n        if (start === -1) {\n          out += line.slice(i);\n          i = line.length;\n        } else {\n          out += `${line.slice(i, start)}    `;\n          i = start + 4;\n          inComment = true;\n        }\n      }\n      return out;\n    },\n    close() {\n      inComment = false;\n    },\n    get inComment() {\n      return inComment;\n    },\n  };\n}\n\n// CommonMark's HTML-block type-6 list (fixed by the spec, not a drifting enumeration): a line\n// starting with an open or close tag of one of these, at column 0 (leading whitespace/`>`\n// blockquote markers tolerated), opens a block that swallows following lines -- tags, links,\n// and comments alike -- until a blank line closes it.\nconst HTML_BLOCK_TAGS = new Set([\n  'address',\n  'article',\n  'aside',\n  'base',\n  'basefont',\n  'blockquote',\n  'body',\n  'caption',\n  'center',\n  'col',\n  'colgroup',\n  'dd',\n  'details',\n  'dialog',\n  'dir',\n  'div',\n  'dl',\n  'dt',\n  'fieldset',\n  'figcaption',\n  'figure',\n  'footer',\n  'form',\n  'frame',\n  'frameset',\n  'h1',\n  'h2',\n  'h3',\n  'h4',\n  'h5',\n  'h6',\n  'head',\n  'header',\n  'hr',\n  'html',\n  'iframe',\n  'legend',\n  'li',\n  'link',\n  'main',\n  'menu',\n  'menuitem',\n  'nav',\n  'noframes',\n  'ol',\n  'optgroup',\n  'option',\n  'p',\n  'param',\n  'search',\n  'section',\n  'summary',\n  'table',\n  'tbody',\n  'td',\n  'tfoot',\n  'th',\n  'thead',\n  'title',\n  'tr',\n  'track',\n  'ul',\n]);\n// Type-1 blocks (script/pre/style/textarea): closes on the line holding the matching end tag,\n// not on a blank line, and that line is the last one swallowed.\nconst HTML_PRE_TAGS = new Set(['script', 'pre', 'style', 'textarea']);\nconst HTML_PRE_CLOSE_RE = /<\\/(?:script|pre|style|textarea)>/i; // any of the four closers ends the block, not only the tag that opened it\n// An opening or closing tag at column 0, tag name captured for the lookups above.\nconst HTML_BLOCK_OPEN_RE = /^<\\/?([a-zA-Z][a-zA-Z0-9]*)(?:[ \\t]|\\/?>|$)/;\nconst BLANK_LINE_RE = /^[ \\t>]*$/; // whitespace/blockquote-markers only -- a paragraph break, and what ends a type-6 block\n\n// Body with fenced-code, HTML-block, and <!-- --> comment regions replaced by spaces (newlines\n// preserved, so line numbers and offsets survive) plus inline code spans masked. The single\n// region masker for tags/links: fence and HTML-block delimiter+content lines are opaque, an\n// unclosed comment dies at its containing block's end (or the next blank line at paragraph\n// level -- never silently to end of file), and a closed comment masks exactly its span.\n// Bodies without a backtick/tilde/`<` fast-path out untouched.\nexport function maskRegions(body: string): string {\n  if (!/[`~<]/.test(body)) return body;\n  const fence = fenceTracker();\n  const comment = commentTracker();\n  let inHtmlBlock = false;\n  let htmlBlockClose: RegExp | null = null; // set while inside a type-1 (script/pre/style/textarea) block\n  return body\n    .split('\\n')\n    .map((line) => {\n      if (inHtmlBlock) {\n        if (htmlBlockClose) {\n          if (htmlBlockClose.test(line)) {\n            inHtmlBlock = false;\n            htmlBlockClose = null;\n          }\n          return ' '.repeat(line.length);\n        }\n        if (BLANK_LINE_RE.test(line)) {\n          // Block content never feeds the comment tracker, so an inner <!-- was only ever\n          // masked text; nothing to close here.\n          inHtmlBlock = false;\n          return line;\n        }\n        return ' '.repeat(line.length);\n      }\n      if (!comment.inComment) {\n        const isDelim = fence.feed(line);\n        if (isDelim || fence.inFence) return ' '.repeat(line.length);\n      }\n      const uncommented = comment.inComment || line.includes('<!--') ? comment.mask(line) : line;\n      if (comment.inComment && BLANK_LINE_RE.test(line)) comment.close(); // paragraph-level: dies at the next blank line\n      // Indented or blockquoted HTML blocks still swallow their content in Obsidian, so the\n      // opener test runs after stripping leading whitespace and > markers.\n      const stripped = uncommented.replace(/^[ \\t>]*/, '');\n      if (stripped[0] === '<') {\n        const openMatch = HTML_BLOCK_OPEN_RE.exec(stripped);\n        if (openMatch) {\n          const tagName = openMatch[1].toLowerCase();\n          const isClosingTag = stripped[1] === '/';\n          if (!isClosingTag && HTML_PRE_TAGS.has(tagName)) {\n            // The end condition can be met on the opening line itself: <script>x</script>\n            // is a one-line block and the next line parses (Obsidian-verified).\n            if (!HTML_PRE_CLOSE_RE.test(stripped.slice(openMatch[0].length))) {\n              inHtmlBlock = true;\n              htmlBlockClose = HTML_PRE_CLOSE_RE;\n            }\n            return ' '.repeat(line.length);\n          }\n          if (HTML_BLOCK_TAGS.has(tagName)) {\n            inHtmlBlock = true;\n            return ' '.repeat(line.length);\n          }\n        }\n      }\n      // Inline code spans hide their content too: `[[x]]` in prose is not a link.\n      return uncommented.includes('`') ? maskCodeSpans(uncommented) : uncommented;\n    })\n    .join('\\n');\n}\n\n// A code span opens on a run of N backticks and closes at the next run of exactly N -- a\n// shorter or longer run in between is literal text, not a delimiter (CommonMark code spans).\n// Masked with spaces so column positions and tag-boundary whitespace are unaffected.\nexport function maskCodeSpans(line: string): string {\n  let out = '';\n  let i = 0;\n  while (i < line.length) {\n    if (line[i] !== '`') {\n      out += line[i];\n      i++;\n      continue;\n    }\n    let j = i;\n    while (line[j] === '`') j++;\n    const n = j - i;\n    let k = j;\n    let closeStart = -1;\n    let closeEnd = -1;\n    while (k < line.length) {\n      if (line[k] !== '`') {\n        k++;\n        continue;\n      }\n      let m = k;\n      while (line[m] === '`') m++;\n      if (m - k === n) {\n        closeStart = k;\n        closeEnd = m;\n        break;\n      }\n      k = m;\n    }\n    if (closeStart >= 0) {\n      out += ' '.repeat(closeEnd - i);\n      i = closeEnd;\n    } else {\n      out += line.slice(i, j);\n      i = j;\n    }\n  }\n  return out;\n}\n"],"names":["BACKTICK_OPEN","TILDE_OPEN","fenceTracker","inFence","fenceChar","fenceLen","feed","line","bt","exec","includes","length","td","run","m","trim","commentTracker","inComment","mask","out","i","end","indexOf","repeat","start","slice","close","HTML_BLOCK_TAGS","Set","HTML_PRE_TAGS","HTML_PRE_CLOSE_RE","HTML_BLOCK_OPEN_RE","BLANK_LINE_RE","maskRegions","body","test","fence","comment","inHtmlBlock","htmlBlockClose","split","map","isDelim","uncommented","stripped","replace","openMatch","tagName","toLowerCase","isClosingTag","has","maskCodeSpans","join","j","n","k","closeStart","closeEnd"],"mappings":"AAAA,+FAA+F;AAC/F,8FAA8F;AAC9F,wGAAwG;AACxG,yFAAyF;AACzF,+DAA+D;AAO/D,MAAMA,gBAAgB;AACtB,MAAMC,aAAa;AAEnB,OAAO,SAASC;IACd,IAAIC,UAAU;IACd,IAAIC,YAAY;IAChB,IAAIC,WAAW;IACf,OAAO;QACLC,MAAKC,IAAY;YACf,IAAI,CAACJ,SAAS;gBACZ,MAAMK,KAAKR,cAAcS,IAAI,CAACF;gBAC9B,IAAIC,MAAM,CAACA,EAAE,CAAC,EAAE,CAACE,QAAQ,CAAC,MAAM;oBAC9BP,UAAU;oBACVC,YAAY;oBACZC,WAAWG,EAAE,CAAC,EAAE,CAACG,MAAM;oBACvB,OAAO;gBACT;gBACA,MAAMC,KAAKX,WAAWQ,IAAI,CAACF;gBAC3B,IAAIK,IAAI;oBACNT,UAAU;oBACVC,YAAY;oBACZC,WAAWO,EAAE,CAAC,EAAE,CAACD,MAAM;oBACvB,OAAO;gBACT;gBACA,OAAO;YACT;YACA,MAAME,MAAMT,cAAc,MAAM,eAAe;YAC/C,MAAMU,IAAID,IAAIJ,IAAI,CAACF;YACnB,IAAIO,KAAKA,CAAC,CAAC,EAAE,CAACH,MAAM,IAAIN,YAAYS,CAAC,CAAC,EAAE,CAACC,IAAI,OAAO,IAAI;gBACtDZ,UAAU;gBACV,OAAO;YACT;YACA,OAAO;QACT;QACA,IAAIA;YACF,OAAOA;QACT;IACF;AACF;AAaA,OAAO,SAASa;IACd,IAAIC,YAAY;IAChB,OAAO;QACLC,MAAKX,IAAY;YACf,IAAIY,MAAM;YACV,IAAIC,IAAI;YACR,MAAOA,IAAIb,KAAKI,MAAM,CAAE;gBACtB,IAAIM,WAAW;oBACb,MAAMI,MAAMd,KAAKe,OAAO,CAAC,OAAOF;oBAChC,IAAIC,QAAQ,CAAC,GAAG;wBACdF,OAAO,IAAII,MAAM,CAAChB,KAAKI,MAAM,GAAGS;wBAChCA,IAAIb,KAAKI,MAAM;oBACjB,OAAO;wBACLQ,OAAO,IAAII,MAAM,CAACF,MAAM,IAAID;wBAC5BA,IAAIC,MAAM;wBACVJ,YAAY;oBACd;oBACA;gBACF;gBACA,MAAMO,QAAQjB,KAAKe,OAAO,CAAC,QAAQF;gBACnC,IAAII,UAAU,CAAC,GAAG;oBAChBL,OAAOZ,KAAKkB,KAAK,CAACL;oBAClBA,IAAIb,KAAKI,MAAM;gBACjB,OAAO;oBACLQ,OAAO,GAAGZ,KAAKkB,KAAK,CAACL,GAAGI,OAAO,IAAI,CAAC;oBACpCJ,IAAII,QAAQ;oBACZP,YAAY;gBACd;YACF;YACA,OAAOE;QACT;QACAO;YACET,YAAY;QACd;QACA,IAAIA;YACF,OAAOA;QACT;IACF;AACF;AAEA,8FAA8F;AAC9F,0FAA0F;AAC1F,6FAA6F;AAC7F,sDAAsD;AACtD,MAAMU,kBAAkB,IAAIC,IAAI;IAC9B;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;CACD;AACD,8FAA8F;AAC9F,gEAAgE;AAChE,MAAMC,gBAAgB,IAAID,IAAI;IAAC;IAAU;IAAO;IAAS;CAAW;AACpE,MAAME,oBAAoB,sCAAsC,0EAA0E;AAC1I,kFAAkF;AAClF,MAAMC,qBAAqB;AAC3B,MAAMC,gBAAgB,aAAa,wFAAwF;AAE3H,+FAA+F;AAC/F,4FAA4F;AAC5F,4FAA4F;AAC5F,2FAA2F;AAC3F,wFAAwF;AACxF,+DAA+D;AAC/D,OAAO,SAASC,YAAYC,IAAY;IACtC,IAAI,CAAC,QAAQC,IAAI,CAACD,OAAO,OAAOA;IAChC,MAAME,QAAQlC;IACd,MAAMmC,UAAUrB;IAChB,IAAIsB,cAAc;IAClB,IAAIC,iBAAgC,MAAM,8DAA8D;IACxG,OAAOL,KACJM,KAAK,CAAC,MACNC,GAAG,CAAC,CAAClC;QACJ,IAAI+B,aAAa;YACf,IAAIC,gBAAgB;gBAClB,IAAIA,eAAeJ,IAAI,CAAC5B,OAAO;oBAC7B+B,cAAc;oBACdC,iBAAiB;gBACnB;gBACA,OAAO,IAAIhB,MAAM,CAAChB,KAAKI,MAAM;YAC/B;YACA,IAAIqB,cAAcG,IAAI,CAAC5B,OAAO;gBAC5B,gFAAgF;gBAChF,sCAAsC;gBACtC+B,cAAc;gBACd,OAAO/B;YACT;YACA,OAAO,IAAIgB,MAAM,CAAChB,KAAKI,MAAM;QAC/B;QACA,IAAI,CAAC0B,QAAQpB,SAAS,EAAE;YACtB,MAAMyB,UAAUN,MAAM9B,IAAI,CAACC;YAC3B,IAAImC,WAAWN,MAAMjC,OAAO,EAAE,OAAO,IAAIoB,MAAM,CAAChB,KAAKI,MAAM;QAC7D;QACA,MAAMgC,cAAcN,QAAQpB,SAAS,IAAIV,KAAKG,QAAQ,CAAC,UAAU2B,QAAQnB,IAAI,CAACX,QAAQA;QACtF,IAAI8B,QAAQpB,SAAS,IAAIe,cAAcG,IAAI,CAAC5B,OAAO8B,QAAQX,KAAK,IAAI,+CAA+C;QACnH,sFAAsF;QACtF,qEAAqE;QACrE,MAAMkB,WAAWD,YAAYE,OAAO,CAAC,YAAY;QACjD,IAAID,QAAQ,CAAC,EAAE,KAAK,KAAK;YACvB,MAAME,YAAYf,mBAAmBtB,IAAI,CAACmC;YAC1C,IAAIE,WAAW;gBACb,MAAMC,UAAUD,SAAS,CAAC,EAAE,CAACE,WAAW;gBACxC,MAAMC,eAAeL,QAAQ,CAAC,EAAE,KAAK;gBACrC,IAAI,CAACK,gBAAgBpB,cAAcqB,GAAG,CAACH,UAAU;oBAC/C,8EAA8E;oBAC9E,oEAAoE;oBACpE,IAAI,CAACjB,kBAAkBK,IAAI,CAACS,SAASnB,KAAK,CAACqB,SAAS,CAAC,EAAE,CAACnC,MAAM,IAAI;wBAChE2B,cAAc;wBACdC,iBAAiBT;oBACnB;oBACA,OAAO,IAAIP,MAAM,CAAChB,KAAKI,MAAM;gBAC/B;gBACA,IAAIgB,gBAAgBuB,GAAG,CAACH,UAAU;oBAChCT,cAAc;oBACd,OAAO,IAAIf,MAAM,CAAChB,KAAKI,MAAM;gBAC/B;YACF;QACF;QACA,4EAA4E;QAC5E,OAAOgC,YAAYjC,QAAQ,CAAC,OAAOyC,cAAcR,eAAeA;IAClE,GACCS,IAAI,CAAC;AACV;AAEA,yFAAyF;AACzF,6FAA6F;AAC7F,qFAAqF;AACrF,OAAO,SAASD,cAAc5C,IAAY;IACxC,IAAIY,MAAM;IACV,IAAIC,IAAI;IACR,MAAOA,IAAIb,KAAKI,MAAM,CAAE;QACtB,IAAIJ,IAAI,CAACa,EAAE,KAAK,KAAK;YACnBD,OAAOZ,IAAI,CAACa,EAAE;YACdA;YACA;QACF;QACA,IAAIiC,IAAIjC;QACR,MAAOb,IAAI,CAAC8C,EAAE,KAAK,IAAKA;QACxB,MAAMC,IAAID,IAAIjC;QACd,IAAImC,IAAIF;QACR,IAAIG,aAAa,CAAC;QAClB,IAAIC,WAAW,CAAC;QAChB,MAAOF,IAAIhD,KAAKI,MAAM,CAAE;YACtB,IAAIJ,IAAI,CAACgD,EAAE,KAAK,KAAK;gBACnBA;gBACA;YACF;YACA,IAAIzC,IAAIyC;YACR,MAAOhD,IAAI,CAACO,EAAE,KAAK,IAAKA;YACxB,IAAIA,IAAIyC,MAAMD,GAAG;gBACfE,aAAaD;gBACbE,WAAW3C;gBACX;YACF;YACAyC,IAAIzC;QACN;QACA,IAAI0C,cAAc,GAAG;YACnBrC,OAAO,IAAII,MAAM,CAACkC,WAAWrC;YAC7BA,IAAIqC;QACN,OAAO;YACLtC,OAAOZ,KAAKkB,KAAK,CAACL,GAAGiC;YACrBjC,IAAIiC;QACN;IACF;IACA,OAAOlC;AACT"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/fences.ts"],"sourcesContent":["// Shared line-fence tracker for tags/sections/embed. Opener: >=3 backticks or tildes at column 0, no indent\n// allowance (diverges from CommonMark's up-to-3-space rule). Closer: same char, length >= opener's, trailing spaces only.\n\nexport interface FenceTracker {\n  feed(line: string): boolean; // true iff this line is a fence delimiter (open or close)\n  readonly inFence: boolean;\n}\n\nconst BACKTICK_OPEN = /^(`{3,})(.*)$/;\nconst TILDE_OPEN = /^(~{3,})(.*)$/;\n\nexport function fenceTracker(): FenceTracker {\n  let inFence = false;\n  let fenceChar = '';\n  let fenceLen = 0;\n  return {\n    feed(line: string): boolean {\n      if (!inFence) {\n        const bt = BACKTICK_OPEN.exec(line);\n        if (bt && !bt[2].includes('`')) {\n          inFence = true;\n          fenceChar = '`';\n          fenceLen = bt[1].length;\n          return true;\n        }\n        const td = TILDE_OPEN.exec(line);\n        if (td) {\n          inFence = true;\n          fenceChar = '~';\n          fenceLen = td[1].length;\n          return true;\n        }\n        return false;\n      }\n      const run = fenceChar === '`' ? /^(`+)(.*)$/ : /^(~+)(.*)$/;\n      const m = run.exec(line);\n      if (m && m[1].length >= fenceLen && m[2].trim() === '') {\n        inFence = false;\n        return true;\n      }\n      return false;\n    },\n    get inFence() {\n      return inFence;\n    },\n  };\n}\n\n// Masks <!-- ... --> spans that may cross line boundaries, one line at a time so callers can interleave their own per-line state.\n// An unclosed comment stays open forever unless closed externally; maskRegions() below calls close() at a paragraph blank line or the enclosing HTML block's end.\nexport interface CommentTracker {\n  mask(line: string): string;\n  close(): void; // force-ends an unclosed comment (paragraph blank line, or its HTML block ending)\n  readonly inComment: boolean;\n}\n\nexport function commentTracker(): CommentTracker {\n  let inComment = false;\n  return {\n    mask(line: string): string {\n      let out = '';\n      let i = 0;\n      while (i < line.length) {\n        if (inComment) {\n          const end = line.indexOf('-->', i);\n          if (end === -1) {\n            out += ' '.repeat(line.length - i);\n            i = line.length;\n          } else {\n            out += ' '.repeat(end + 3 - i);\n            i = end + 3;\n            inComment = false;\n          }\n          continue;\n        }\n        const start = line.indexOf('<!--', i);\n        if (start === -1) {\n          out += line.slice(i);\n          i = line.length;\n        } else {\n          out += `${line.slice(i, start)}    `;\n          i = start + 4;\n          inComment = true;\n        }\n      }\n      return out;\n    },\n    close() {\n      inComment = false;\n    },\n    get inComment() {\n      return inComment;\n    },\n  };\n}\n\n// CommonMark's HTML-block type-6 tag list (fixed by the spec). A line opening or closing one of these at\n// column 0 (leading whitespace/`>` tolerated) swallows following lines -- tags, links, comments alike -- until a blank line.\nconst HTML_BLOCK_TAGS = new Set([\n  'address',\n  'article',\n  'aside',\n  'base',\n  'basefont',\n  'blockquote',\n  'body',\n  'caption',\n  'center',\n  'col',\n  'colgroup',\n  'dd',\n  'details',\n  'dialog',\n  'dir',\n  'div',\n  'dl',\n  'dt',\n  'fieldset',\n  'figcaption',\n  'figure',\n  'footer',\n  'form',\n  'frame',\n  'frameset',\n  'h1',\n  'h2',\n  'h3',\n  'h4',\n  'h5',\n  'h6',\n  'head',\n  'header',\n  'hr',\n  'html',\n  'iframe',\n  'legend',\n  'li',\n  'link',\n  'main',\n  'menu',\n  'menuitem',\n  'nav',\n  'noframes',\n  'ol',\n  'optgroup',\n  'option',\n  'p',\n  'param',\n  'search',\n  'section',\n  'summary',\n  'table',\n  'tbody',\n  'td',\n  'tfoot',\n  'th',\n  'thead',\n  'title',\n  'tr',\n  'track',\n  'ul',\n]);\n// Type-1 blocks (script/pre/style/textarea): closes on the line holding the matching end tag,\n// not on a blank line, and that line is the last one swallowed.\nconst HTML_PRE_TAGS = new Set(['script', 'pre', 'style', 'textarea']);\nconst HTML_PRE_CLOSE_RE = /<\\/(?:script|pre|style|textarea)>/i; // any of the four closers ends the block, not only the tag that opened it\n// An opening or closing tag at column 0, tag name captured for the lookups above.\nconst HTML_BLOCK_OPEN_RE = /^<\\/?([a-zA-Z][a-zA-Z0-9]*)(?:[ \\t]|\\/?>|$)/;\nconst BLANK_LINE_RE = /^[ \\t>]*$/; // whitespace/blockquote-markers only -- a paragraph break, and what ends a type-6 block\n\n// Body with fenced-code, HTML-block, and <!-- --> comment regions replaced by spaces (newlines preserved, so line numbers and offsets survive), inline code spans masked too.\n// An unclosed comment dies at its block's end or the next paragraph blank line, never silently to end of file; bodies without a backtick/tilde/`<` fast-path out untouched.\nexport function maskRegions(body: string): string {\n  if (!/[`~<]/.test(body)) return body;\n  const fence = fenceTracker();\n  const comment = commentTracker();\n  let inHtmlBlock = false;\n  let htmlBlockClose: RegExp | null = null; // set while inside a type-1 (script/pre/style/textarea) block\n  return body\n    .split('\\n')\n    .map((line) => {\n      if (inHtmlBlock) {\n        if (htmlBlockClose) {\n          if (htmlBlockClose.test(line)) {\n            inHtmlBlock = false;\n            htmlBlockClose = null;\n          }\n          return ' '.repeat(line.length);\n        }\n        if (BLANK_LINE_RE.test(line)) {\n          // Block content never feeds the comment tracker, so an inner <!-- was only ever\n          // masked text; nothing to close here.\n          inHtmlBlock = false;\n          return line;\n        }\n        return ' '.repeat(line.length);\n      }\n      if (!comment.inComment) {\n        const isDelim = fence.feed(line);\n        if (isDelim || fence.inFence) return ' '.repeat(line.length);\n      }\n      const uncommented = comment.inComment || line.includes('<!--') ? comment.mask(line) : line;\n      if (comment.inComment && BLANK_LINE_RE.test(line)) comment.close(); // paragraph-level: dies at the next blank line\n      // Indented or blockquoted HTML blocks swallow their content in Obsidian too, so the\n      // opener test runs after stripping leading whitespace and > markers.\n      const stripped = uncommented.replace(/^[ \\t>]*/, '');\n      if (stripped[0] === '<') {\n        const openMatch = HTML_BLOCK_OPEN_RE.exec(stripped);\n        if (openMatch) {\n          const tagName = openMatch[1].toLowerCase();\n          const isClosingTag = stripped[1] === '/';\n          if (!isClosingTag && HTML_PRE_TAGS.has(tagName)) {\n            // The end condition can be met on the opening line itself: <script>x</script>\n            // is a one-line block and the next line parses (Obsidian-verified).\n            if (!HTML_PRE_CLOSE_RE.test(stripped.slice(openMatch[0].length))) {\n              inHtmlBlock = true;\n              htmlBlockClose = HTML_PRE_CLOSE_RE;\n            }\n            return ' '.repeat(line.length);\n          }\n          if (HTML_BLOCK_TAGS.has(tagName)) {\n            inHtmlBlock = true;\n            return ' '.repeat(line.length);\n          }\n        }\n      }\n      // Inline code spans hide their content too: `[[x]]` in prose is not a link.\n      return uncommented.includes('`') ? maskCodeSpans(uncommented) : uncommented;\n    })\n    .join('\\n');\n}\n\n// A code span opens on a run of N backticks and closes at the next run of exactly N (CommonMark code spans);\n// a shorter/longer run in between is literal text. Masked with spaces so column positions stay unaffected.\nexport function maskCodeSpans(line: string): string {\n  let out = '';\n  let i = 0;\n  while (i < line.length) {\n    if (line[i] !== '`') {\n      out += line[i];\n      i++;\n      continue;\n    }\n    let j = i;\n    while (line[j] === '`') j++;\n    const n = j - i;\n    let k = j;\n    let closeStart = -1;\n    let closeEnd = -1;\n    while (k < line.length) {\n      if (line[k] !== '`') {\n        k++;\n        continue;\n      }\n      let m = k;\n      while (line[m] === '`') m++;\n      if (m - k === n) {\n        closeStart = k;\n        closeEnd = m;\n        break;\n      }\n      k = m;\n    }\n    if (closeStart >= 0) {\n      out += ' '.repeat(closeEnd - i);\n      i = closeEnd;\n    } else {\n      out += line.slice(i, j);\n      i = j;\n    }\n  }\n  return out;\n}\n"],"names":["BACKTICK_OPEN","TILDE_OPEN","fenceTracker","inFence","fenceChar","fenceLen","feed","line","bt","exec","includes","length","td","run","m","trim","commentTracker","inComment","mask","out","i","end","indexOf","repeat","start","slice","close","HTML_BLOCK_TAGS","Set","HTML_PRE_TAGS","HTML_PRE_CLOSE_RE","HTML_BLOCK_OPEN_RE","BLANK_LINE_RE","maskRegions","body","test","fence","comment","inHtmlBlock","htmlBlockClose","split","map","isDelim","uncommented","stripped","replace","openMatch","tagName","toLowerCase","isClosingTag","has","maskCodeSpans","join","j","n","k","closeStart","closeEnd"],"mappings":"AAAA,4GAA4G;AAC5G,0HAA0H;AAO1H,MAAMA,gBAAgB;AACtB,MAAMC,aAAa;AAEnB,OAAO,SAASC;IACd,IAAIC,UAAU;IACd,IAAIC,YAAY;IAChB,IAAIC,WAAW;IACf,OAAO;QACLC,MAAKC,IAAY;YACf,IAAI,CAACJ,SAAS;gBACZ,MAAMK,KAAKR,cAAcS,IAAI,CAACF;gBAC9B,IAAIC,MAAM,CAACA,EAAE,CAAC,EAAE,CAACE,QAAQ,CAAC,MAAM;oBAC9BP,UAAU;oBACVC,YAAY;oBACZC,WAAWG,EAAE,CAAC,EAAE,CAACG,MAAM;oBACvB,OAAO;gBACT;gBACA,MAAMC,KAAKX,WAAWQ,IAAI,CAACF;gBAC3B,IAAIK,IAAI;oBACNT,UAAU;oBACVC,YAAY;oBACZC,WAAWO,EAAE,CAAC,EAAE,CAACD,MAAM;oBACvB,OAAO;gBACT;gBACA,OAAO;YACT;YACA,MAAME,MAAMT,cAAc,MAAM,eAAe;YAC/C,MAAMU,IAAID,IAAIJ,IAAI,CAACF;YACnB,IAAIO,KAAKA,CAAC,CAAC,EAAE,CAACH,MAAM,IAAIN,YAAYS,CAAC,CAAC,EAAE,CAACC,IAAI,OAAO,IAAI;gBACtDZ,UAAU;gBACV,OAAO;YACT;YACA,OAAO;QACT;QACA,IAAIA;YACF,OAAOA;QACT;IACF;AACF;AAUA,OAAO,SAASa;IACd,IAAIC,YAAY;IAChB,OAAO;QACLC,MAAKX,IAAY;YACf,IAAIY,MAAM;YACV,IAAIC,IAAI;YACR,MAAOA,IAAIb,KAAKI,MAAM,CAAE;gBACtB,IAAIM,WAAW;oBACb,MAAMI,MAAMd,KAAKe,OAAO,CAAC,OAAOF;oBAChC,IAAIC,QAAQ,CAAC,GAAG;wBACdF,OAAO,IAAII,MAAM,CAAChB,KAAKI,MAAM,GAAGS;wBAChCA,IAAIb,KAAKI,MAAM;oBACjB,OAAO;wBACLQ,OAAO,IAAII,MAAM,CAACF,MAAM,IAAID;wBAC5BA,IAAIC,MAAM;wBACVJ,YAAY;oBACd;oBACA;gBACF;gBACA,MAAMO,QAAQjB,KAAKe,OAAO,CAAC,QAAQF;gBACnC,IAAII,UAAU,CAAC,GAAG;oBAChBL,OAAOZ,KAAKkB,KAAK,CAACL;oBAClBA,IAAIb,KAAKI,MAAM;gBACjB,OAAO;oBACLQ,OAAO,GAAGZ,KAAKkB,KAAK,CAACL,GAAGI,OAAO,IAAI,CAAC;oBACpCJ,IAAII,QAAQ;oBACZP,YAAY;gBACd;YACF;YACA,OAAOE;QACT;QACAO;YACET,YAAY;QACd;QACA,IAAIA;YACF,OAAOA;QACT;IACF;AACF;AAEA,yGAAyG;AACzG,6HAA6H;AAC7H,MAAMU,kBAAkB,IAAIC,IAAI;IAC9B;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;CACD;AACD,8FAA8F;AAC9F,gEAAgE;AAChE,MAAMC,gBAAgB,IAAID,IAAI;IAAC;IAAU;IAAO;IAAS;CAAW;AACpE,MAAME,oBAAoB,sCAAsC,0EAA0E;AAC1I,kFAAkF;AAClF,MAAMC,qBAAqB;AAC3B,MAAMC,gBAAgB,aAAa,wFAAwF;AAE3H,8KAA8K;AAC9K,4KAA4K;AAC5K,OAAO,SAASC,YAAYC,IAAY;IACtC,IAAI,CAAC,QAAQC,IAAI,CAACD,OAAO,OAAOA;IAChC,MAAME,QAAQlC;IACd,MAAMmC,UAAUrB;IAChB,IAAIsB,cAAc;IAClB,IAAIC,iBAAgC,MAAM,8DAA8D;IACxG,OAAOL,KACJM,KAAK,CAAC,MACNC,GAAG,CAAC,CAAClC;QACJ,IAAI+B,aAAa;YACf,IAAIC,gBAAgB;gBAClB,IAAIA,eAAeJ,IAAI,CAAC5B,OAAO;oBAC7B+B,cAAc;oBACdC,iBAAiB;gBACnB;gBACA,OAAO,IAAIhB,MAAM,CAAChB,KAAKI,MAAM;YAC/B;YACA,IAAIqB,cAAcG,IAAI,CAAC5B,OAAO;gBAC5B,gFAAgF;gBAChF,sCAAsC;gBACtC+B,cAAc;gBACd,OAAO/B;YACT;YACA,OAAO,IAAIgB,MAAM,CAAChB,KAAKI,MAAM;QAC/B;QACA,IAAI,CAAC0B,QAAQpB,SAAS,EAAE;YACtB,MAAMyB,UAAUN,MAAM9B,IAAI,CAACC;YAC3B,IAAImC,WAAWN,MAAMjC,OAAO,EAAE,OAAO,IAAIoB,MAAM,CAAChB,KAAKI,MAAM;QAC7D;QACA,MAAMgC,cAAcN,QAAQpB,SAAS,IAAIV,KAAKG,QAAQ,CAAC,UAAU2B,QAAQnB,IAAI,CAACX,QAAQA;QACtF,IAAI8B,QAAQpB,SAAS,IAAIe,cAAcG,IAAI,CAAC5B,OAAO8B,QAAQX,KAAK,IAAI,+CAA+C;QACnH,oFAAoF;QACpF,qEAAqE;QACrE,MAAMkB,WAAWD,YAAYE,OAAO,CAAC,YAAY;QACjD,IAAID,QAAQ,CAAC,EAAE,KAAK,KAAK;YACvB,MAAME,YAAYf,mBAAmBtB,IAAI,CAACmC;YAC1C,IAAIE,WAAW;gBACb,MAAMC,UAAUD,SAAS,CAAC,EAAE,CAACE,WAAW;gBACxC,MAAMC,eAAeL,QAAQ,CAAC,EAAE,KAAK;gBACrC,IAAI,CAACK,gBAAgBpB,cAAcqB,GAAG,CAACH,UAAU;oBAC/C,8EAA8E;oBAC9E,oEAAoE;oBACpE,IAAI,CAACjB,kBAAkBK,IAAI,CAACS,SAASnB,KAAK,CAACqB,SAAS,CAAC,EAAE,CAACnC,MAAM,IAAI;wBAChE2B,cAAc;wBACdC,iBAAiBT;oBACnB;oBACA,OAAO,IAAIP,MAAM,CAAChB,KAAKI,MAAM;gBAC/B;gBACA,IAAIgB,gBAAgBuB,GAAG,CAACH,UAAU;oBAChCT,cAAc;oBACd,OAAO,IAAIf,MAAM,CAAChB,KAAKI,MAAM;gBAC/B;YACF;QACF;QACA,4EAA4E;QAC5E,OAAOgC,YAAYjC,QAAQ,CAAC,OAAOyC,cAAcR,eAAeA;IAClE,GACCS,IAAI,CAAC;AACV;AAEA,6GAA6G;AAC7G,2GAA2G;AAC3G,OAAO,SAASD,cAAc5C,IAAY;IACxC,IAAIY,MAAM;IACV,IAAIC,IAAI;IACR,MAAOA,IAAIb,KAAKI,MAAM,CAAE;QACtB,IAAIJ,IAAI,CAACa,EAAE,KAAK,KAAK;YACnBD,OAAOZ,IAAI,CAACa,EAAE;YACdA;YACA;QACF;QACA,IAAIiC,IAAIjC;QACR,MAAOb,IAAI,CAAC8C,EAAE,KAAK,IAAKA;QACxB,MAAMC,IAAID,IAAIjC;QACd,IAAImC,IAAIF;QACR,IAAIG,aAAa,CAAC;QAClB,IAAIC,WAAW,CAAC;QAChB,MAAOF,IAAIhD,KAAKI,MAAM,CAAE;YACtB,IAAIJ,IAAI,CAACgD,EAAE,KAAK,KAAK;gBACnBA;gBACA;YACF;YACA,IAAIzC,IAAIyC;YACR,MAAOhD,IAAI,CAACO,EAAE,KAAK,IAAKA;YACxB,IAAIA,IAAIyC,MAAMD,GAAG;gBACfE,aAAaD;gBACbE,WAAW3C;gBACX;YACF;YACAyC,IAAIzC;QACN;QACA,IAAI0C,cAAc,GAAG;YACnBrC,OAAO,IAAII,MAAM,CAACkC,WAAWrC;YAC7BA,IAAIqC;QACN,OAAO;YACLtC,OAAOZ,KAAKkB,KAAK,CAACL,GAAGiC;YACrBjC,IAAIiC;QACN;IACF;IACA,OAAOlC;AACT"}

package/dist/esm/features/links.js

@@ -1,21 +1,17 @@
 import posix from 'node:path/posix';
 import { maskRegions } from './fences.js';
-// links(src, target, target_base, dst, embed): target as written, target_base its baseKey
-// (indexed, for the incremental resolve below), dst the resolved path or NULL (a
-// queryable dead link), embed whether it's `![[x]]`/`![](x.md)` rather than `[[x]]`/`[](x.md)`
-// -- Obsidian's grain: a target both linked and embedded in the same note is two rows.
-// Wikilinks ([[target]], [[target#anchor|alias]], embeds), internal markdown links, and
-// frontmatter values that are exactly a wikilink ("[[X]]"; mid-string and ![[...]] forms are
-// not links there -- Obsidian's rule, probe-verified).
-// One rule for [[inner]] text wherever it appears: alias stripped after |, a leading #
-// keeps the anchor as the target (a same-note self-link needs a heading name), otherwise
-// the anchor splits off. Null = not a link.
+// links(src, target, target_base, dst, embed): target as written, target_base its indexed baseKey, dst the resolved path or NULL (a queryable dead link).
+// embed distinguishes `![[x]]`/`![](x.md)` from `[[x]]`/`[](x.md)` -- Obsidian's grain: a target both linked and embedded in one note is two rows.
+// Alias stripped after |; a leading # keeps the anchor as the target (self-link needs a
+// heading name), else the anchor splits off. Null = not a link.
 function parseWikilinkInner(inner) {
     const beforeAlias = inner.split('|')[0].trim();
     if (beforeAlias.startsWith('#')) return beforeAlias.length > 1 ? beforeAlias : null;
     const target = beforeAlias.split('#')[0].trim();
     return target || null;
 }
+// A frontmatter value counts only when it is exactly a wikilink ("[[X]]") -- mid-string and
+// ![[...]] forms don't count there, per Obsidian's rule (probe-verified).
 function extract(_raw, rawBody, _search, data) {
     const body = /\[\[|\]\(/.test(rawBody) ? maskRegions(rawBody) : rawBody;
     const seen = new Set();
@@ -36,12 +32,8 @@ function extract(_raw, rawBody, _search, data) {
         const target = parseWikilinkInner(m[1]);
         if (target) add(target, body[((_m_index = m.index) !== null && _m_index !== void 0 ? _m_index : 0) - 1] === '!');
     }
-    // Any internal destination, not only .md-suffixed: Obsidian gives markdown links full
-    // linkpath resolution, so \](Zektor) resolves like [[Zektor]] and \](#anchor) is a
-    // self-edge. External URLs (a scheme anywhere survives the malformed double-paren case)
-    // and titled links (dest stops at whitespace) are skipped.
-    // A space in the destination is only valid ahead of a quoted title; a domain-shaped first
-    // segment (www.example.com/...) is external even without a scheme.
+    // Any internal destination, not only .md-suffixed: \](Zektor) resolves like [[Zektor]] and
+    // \](#anchor) is a self-edge. External URLs and titled links (dest stops at whitespace) skip.
     for (const m of body.matchAll(/(!)?\[[^\]]*\]\(((?:[^()\s]|\([^()\s]*\))+)(?:\s+(?:"[^)]*"|'[^)]*'))?\)/g)){
         const dest = m[2].trim();
         // External: a URI scheme prefix (mailto:, tel:, data:, https:...), protocol-relative
@@ -70,9 +62,8 @@ function cleanTarget(target) {
 function baseKey(path) {
     return posix.basename(path).replace(/\.md$/i, '').toLowerCase();
 }
-// Obsidian-style: exact relative path (with/without .md), path relative to the linking
-// note's directory, then basename match (shortest path wins on ties -- verified against a
-// cold-loaded Obsidian vault on 6+ real collision pairs; byBase's lists are pre-sorted that way).
+// Obsidian-style: exact relative path (with/without .md), path relative to the linking note's
+// directory, then basename match (shortest path wins ties, verified against a cold-loaded vault).
 function resolveTarget(src, target, pathSet, byBase) {
     // [[#Heading]]: a same-note anchor, always the note itself -- never depends on other files.
     if (target.startsWith('#')) return src;
@@ -107,9 +98,8 @@ function buildByBase(files) {
     }
     return byBase;
 }
-// resolveTarget applied to `rows`, returning just the rows whose dst actually changed -- the
-// caller writes those in one runBatch call, keyed on (src, target) so a link/embed sibling
-// pair sharing one resolution rewrites both.
+// resolveTarget applied to `rows`, returning only the rows whose dst changed, so the caller
+// can write them in one runBatch call.
 function resolveRows(rows, pathSet, byBase) {
     const changed = [];
     for (const row of rows){
@@ -151,9 +141,8 @@ async function selectIn(db, column, keys) {
     }
     return rows;
 }
-// Re-resolve only what this reconcile could have touched: re-stored sources, links whose
-// target basename matches an added/vanished file, and links whose dst just vanished. All
-// three are indexed lookups, so this never reads the whole table.
+// Re-resolves only what this reconcile could have touched: re-stored sources, links whose
+// target basename matches an added/vanished file, and links whose dst vanished -- all indexed.
 async function resolveIncremental(db, delta) {
     const pathSet = new Set(delta.files.map((f)=>f.relPath));
     const byBase = buildByBase(delta.files);
@@ -176,15 +165,8 @@ async function resolveIncremental(db, delta) {
     await writeResolved(db, changed);
     return changed.length > 0;
 }
-// One edge per row, as 0.12's grain always had it -- two written targets resolving to one
-// dst stay two edges (a weight the fusion evals were gated on). The only exclusion is an
-// embed whose exact (src, target) also exists as a link: that second row is new with the
-// embed grain and would double an edge that used to be one row. The NOT EXISTS probes the
-// primary key and fires only for embed rows; both branches still scan the table.
-// Self-edges (same-note anchors) are excluded here so PageRank mass is not self-recycled --
-// Obsidian's own graph view hides self-loops too. The rows themselves stay in the table for
-// backlinks/peek. Exported (not just linkEdges below) so commands/signals.ts's async caller,
-// which only has the Store's prepare(), can run the identical statement itself.
+// One edge per (src, target); an embed row is excluded only when the same pair also exists as
+// a link. Self-edges are dropped so PageRank mass is not self-recycled.
 export const LINK_EDGES_SQL = `SELECT src, dst FROM links WHERE dst IS NOT NULL AND embed = 0 AND src != dst
     UNION ALL
     SELECT src, dst FROM links l WHERE dst IS NOT NULL AND embed = 1 AND src != dst
@@ -200,10 +182,8 @@ export async function linkEdges(db) {
     const stmt = await db.prepare(LINK_EDGES_SQL);
     return toEdges(await stmt.all());
 }
-// remove()/store() have already deleted rows by the time afterReconcile runs, so this records
-// whether any reconcile pass deleted at least one link row -- whether a file vanished, or a
-// reparse's stale rows shrank. dstChanged alone misses the reparsed-to-zero-links case (no
-// surviving rows to re-resolve). Keyed on the delta object, so the state dies with the reconcile.
+// Records whether remove()/store() deleted any link row, since afterReconcile runs after both
+// and dstChanged alone misses the reparsed-to-zero-links case. Keyed on delta so state dies with it.
 const removedWithLinks = new WeakMap();
 export const links = {
     name: 'links',
@@ -213,10 +193,8 @@ export const links = {
         await db.exec('CREATE INDEX IF NOT EXISTS links_target_base ON links(target_base)');
     },
     extract,
-    // Reparsed docs are diffed in store() below instead of wiped here: deleting and re-inserting
-    // resets dst to NULL, which made every touch-only reparse read as an edge change and
-    // recompute PageRank -- measured as most of the remaining update cost. Only genuinely
-    // vanished files clear here.
+    // Reparsed docs are diffed in store() instead of wiped here: delete-and-reinsert resets dst
+    // to NULL, making every touch-only reparse look like an edge change and recompute PageRank.
     async remove (db, paths, delta) {
         const vanishedSet = new Set(delta.vanished);
         const vanished = paths.filter((p)=>vanishedSet.has(p));
@@ -234,9 +212,8 @@ export const links = {
             ]));
         const addedSet = new Set(delta.added);
         const reparsedExisting = docs.map((d)=>d.path).filter((p)=>!addedSet.has(p));
-        // Stale rows (existing target/embed pairs the new parse no longer contains) are real edge
-        // removals; diffed in JS from one bulk SELECT rather than a per-file NOT IN statement, so
-        // this whole pass costs one query plus one batched DELETE regardless of file count.
+        // Stale rows (target/embed pairs missing from the new parse) are diffed in JS from one
+        // bulk SELECT, so this pass costs one query plus one batched DELETE regardless of file count.
         const existingRows = await selectIn(db, 'src', reparsedExisting);
         const staleRows = existingRows.filter((row)=>{
             var _byPath_get;
@@ -251,10 +228,8 @@ export const links = {
                 ]));
             removedWithLinks.set(delta, true);
         }
-        // Upsert preserves dst on surviving rows, so an unchanged link resolves to the same value
-        // and reports no change; only genuinely new rows start at NULL. A same-note anchor's dst is
-        // always its own src, never another file's basename, so it gets no target_base -- SQL's
-        // `IN` never matches NULL, keeping it out of resolveIncremental's target_base lookup.
+        // Upsert preserves dst on surviving rows; only new rows start at NULL. A same-note anchor
+        // gets no target_base, so SQL's `IN` excludes it from resolveIncremental's lookup.
         const upsertRows = [];
         for (const [path, targets] of byPath){
             for (const { target, embed } of targets)upsertRows.push([

package/dist/esm/features/links.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/links.ts"],"sourcesContent":["import posix from 'node:path/posix';\nimport type { FileStat } from '../scan/index.ts';\nimport type { Connection } from '../store/types.ts';\nimport { maskRegions } from './fences.ts';\nimport type { ExtractedDoc, Feature, ReconcileDelta } from './types.ts';\n\n// links(src, target, target_base, dst, embed): target as written, target_base its baseKey\n// (indexed, for the incremental resolve below), dst the resolved path or NULL (a\n// queryable dead link), embed whether it's `![[x]]`/`![](x.md)` rather than `[[x]]`/`[](x.md)`\n// -- Obsidian's grain: a target both linked and embedded in the same note is two rows.\n\n// Wikilinks ([[target]], [[target#anchor|alias]], embeds), internal markdown links, and\n// frontmatter values that are exactly a wikilink (\"[[X]]\"; mid-string and ![[...]] forms are\n// not links there -- Obsidian's rule, probe-verified).\n// One rule for [[inner]] text wherever it appears: alias stripped after |, a leading #\n// keeps the anchor as the target (a same-note self-link needs a heading name), otherwise\n// the anchor splits off. Null = not a link.\nfunction parseWikilinkInner(inner: string): string | null {\n  const beforeAlias = inner.split('|')[0].trim();\n  if (beforeAlias.startsWith('#')) return beforeAlias.length > 1 ? beforeAlias : null;\n  const target = beforeAlias.split('#')[0].trim();\n  return target || null;\n}\n\nfunction extract(_raw: string, rawBody: string, _search?: { title: string; summary: string }, data?: Record<string, unknown>): Array<{ target: string; embed: boolean }> {\n  const body = /\\[\\[|\\]\\(/.test(rawBody) ? maskRegions(rawBody) : rawBody;\n  const seen = new Set<string>();\n  const results: Array<{ target: string; embed: boolean }> = [];\n  const add = (target: string, embed: boolean) => {\n    const key = `${target}\\0${embed ? '1' : '0'}`;\n    if (seen.has(key)) return;\n    seen.add(key);\n    results.push({ target, embed });\n  };\n  // To the first ]], as Obsidian parses it, so a heading or alias holding a lone ] still\n  // matches; anchor and alias split off in parseWikilinkInner.\n  for (const m of body.matchAll(/\\[\\[(.*?)\\]\\]/g)) {\n    const target = parseWikilinkInner(m[1]);\n    if (target) add(target, body[(m.index ?? 0) - 1] === '!');\n  }\n  // Any internal destination, not only .md-suffixed: Obsidian gives markdown links full\n  // linkpath resolution, so \\](Zektor) resolves like [[Zektor]] and \\](#anchor) is a\n  // self-edge. External URLs (a scheme anywhere survives the malformed double-paren case)\n  // and titled links (dest stops at whitespace) are skipped.\n  // A space in the destination is only valid ahead of a quoted title; a domain-shaped first\n  // segment (www.example.com/...) is external even without a scheme.\n  for (const m of body.matchAll(/(!)?\\[[^\\]]*\\]\\(((?:[^()\\s]|\\([^()\\s]*\\))+)(?:\\s+(?:\"[^)]*\"|'[^)]*'))?\\)/g)) {\n    const dest = m[2].trim();\n    // External: a URI scheme prefix (mailto:, tel:, data:, https:...), protocol-relative\n    // //host, a www. shorthand, or a scheme anywhere (the malformed double-paren case).\n    if (!dest || dest.includes('://')) continue;\n    if (/^([a-z][a-z0-9+.-]*:|\\/\\/|www\\.)/i.test(dest)) continue;\n    const target = dest.startsWith('#') ? dest : dest.split('#')[0];\n    if (target) add(target, m[1] === '!');\n  }\n  for (const value of Object.values(data ?? {})) {\n    for (const item of Array.isArray(value) ? value : [value]) {\n      if (typeof item !== 'string') continue;\n      const m = /^\\[\\[(.*)\\]\\]$/.exec(item.trim());\n      if (!m) continue;\n      const target = parseWikilinkInner(m[1]);\n      if (target) add(target, false);\n    }\n  }\n  return results;\n}\n\nfunction cleanTarget(target: string): string {\n  return target.replace(/\\\\/g, '/').replace(/^\\.\\//, '');\n}\n\nfunction baseKey(path: string): string {\n  return posix.basename(path).replace(/\\.md$/i, '').toLowerCase();\n}\n\n// Obsidian-style: exact relative path (with/without .md), path relative to the linking\n// note's directory, then basename match (shortest path wins on ties -- verified against a\n// cold-loaded Obsidian vault on 6+ real collision pairs; byBase's lists are pre-sorted that way).\nfunction resolveTarget(src: string, target: string, pathSet: Set<string>, byBase: Map<string, string[]>): string | null {\n  // [[#Heading]]: a same-note anchor, always the note itself -- never depends on other files.\n  if (target.startsWith('#')) return src;\n  const clean = cleanTarget(target);\n  const fromSrc = posix.normalize(posix.join(posix.dirname(src), clean));\n  for (const candidate of [clean, `${clean}.md`, fromSrc, `${fromSrc}.md`]) {\n    if (pathSet.has(candidate)) return candidate;\n  }\n  const candidates = byBase.get(baseKey(clean));\n  if (!candidates) return null;\n  // The linking note itself wins a basename collision (Obsidian resolves to self before the\n  // shortest-path rule -- cold-load verified on the hub corpus).\n  return candidates.includes(src) ? src : candidates[0];\n}\n\n// Shortest path wins a basename collision; equal lengths fall back to lexicographic, our own\n// deterministic tiebreak for a case Obsidian itself leaves registration-order-dependent.\nfunction buildByBase(files: FileStat[]): Map<string, string[]> {\n  const byBase = new Map<string, string[]>();\n  const paths = files.map((f) => f.relPath).sort((a, b) => a.length - b.length || (a < b ? -1 : a > b ? 1 : 0));\n  for (const path of paths) {\n    const key = baseKey(path);\n    const list = byBase.get(key);\n    if (list) list.push(path);\n    else byBase.set(key, [path]);\n  }\n  return byBase;\n}\n\ninterface LinkRow {\n  src: string;\n  target: string;\n  dst: string | null;\n  embed: number;\n}\n\n// resolveTarget applied to `rows`, returning just the rows whose dst actually changed -- the\n// caller writes those in one runBatch call, keyed on (src, target) so a link/embed sibling\n// pair sharing one resolution rewrites both.\nfunction resolveRows(rows: LinkRow[], pathSet: Set<string>, byBase: Map<string, string[]>): LinkRow[] {\n  const changed: LinkRow[] = [];\n  for (const row of rows) {\n    const dst = resolveTarget(row.src, row.target, pathSet, byBase);\n    if (dst !== row.dst) changed.push({ ...row, dst });\n  }\n  return changed;\n}\n\nasync function writeResolved(db: Connection, rows: LinkRow[]): Promise<void> {\n  if (rows.length === 0) return;\n  await db.runBatch(\n    'UPDATE links SET dst = ? WHERE src = ? AND target = ?',\n    rows.map((r) => [r.dst, r.src, r.target])\n  );\n}\n\n// A new or deleted file can change any note's resolution, so re-resolve the whole table.\n// Fallback for cold builds and large deltas -- see afterReconcile's threshold.\nasync function resolveAll(db: Connection, files: FileStat[]): Promise<boolean> {\n  const pathSet = new Set(files.map((f) => f.relPath));\n  const byBase = buildByBase(files);\n  const stmt = await db.prepare('SELECT src, target, dst, embed FROM links');\n  const rows = (await stmt.all()) as unknown as LinkRow[];\n  const changed = resolveRows(rows, pathSet, byBase);\n  await writeResolved(db, changed);\n  return changed.length > 0;\n}\n\n// Chunked so a large-but-under-threshold delta never exceeds SQLite's bound-variable limit\n// (SQLITE_MAX_VARIABLE_NUMBER, 32766); harmless headroom on DuckDB, which has no such cap.\nasync function selectIn(db: Connection, column: string, keys: string[]): Promise<LinkRow[]> {\n  const rows: LinkRow[] = [];\n  for (let i = 0; i < keys.length; i += 500) {\n    const chunk = keys.slice(i, i + 500);\n    const stmt = await db.prepare(`SELECT src, target, dst, embed FROM links WHERE ${column} IN (${chunk.map(() => '?').join(', ')})`);\n    rows.push(...((await stmt.all(...chunk)) as unknown as LinkRow[]));\n  }\n  return rows;\n}\n\n// Re-resolve only what this reconcile could have touched: re-stored sources, links whose\n// target basename matches an added/vanished file, and links whose dst just vanished. All\n// three are indexed lookups, so this never reads the whole table.\nasync function resolveIncremental(db: Connection, delta: ReconcileDelta): Promise<boolean> {\n  const pathSet = new Set(delta.files.map((f) => f.relPath));\n  const byBase = buildByBase(delta.files);\n\n  const changedBasenames = new Set<string>();\n  for (const p of delta.added) changedBasenames.add(baseKey(p));\n  for (const p of delta.vanished) changedBasenames.add(baseKey(p));\n\n  // \\0 can't appear in a path or target, so the key never collides; a space can.\n  const candidates = new Map<string, LinkRow>();\n  const collect = (rows: LinkRow[]) => {\n    for (const row of rows) candidates.set(`${row.src}\\0${row.target}\\0${row.embed}`, row);\n  };\n  collect(await selectIn(db, 'src', delta.reparsed));\n  collect(await selectIn(db, 'target_base', [...changedBasenames]));\n  collect(await selectIn(db, 'dst', delta.vanished));\n\n  const changed = resolveRows([...candidates.values()], pathSet, byBase);\n  await writeResolved(db, changed);\n  return changed.length > 0;\n}\n\n// One edge per row, as 0.12's grain always had it -- two written targets resolving to one\n// dst stay two edges (a weight the fusion evals were gated on). The only exclusion is an\n// embed whose exact (src, target) also exists as a link: that second row is new with the\n// embed grain and would double an edge that used to be one row. The NOT EXISTS probes the\n// primary key and fires only for embed rows; both branches still scan the table.\n// Self-edges (same-note anchors) are excluded here so PageRank mass is not self-recycled --\n// Obsidian's own graph view hides self-loops too. The rows themselves stay in the table for\n// backlinks/peek. Exported (not just linkEdges below) so commands/signals.ts's async caller,\n// which only has the Store's prepare(), can run the identical statement itself.\nexport const LINK_EDGES_SQL = `SELECT src, dst FROM links WHERE dst IS NOT NULL AND embed = 0 AND src != dst\n    UNION ALL\n    SELECT src, dst FROM links l WHERE dst IS NOT NULL AND embed = 1 AND src != dst\n      AND NOT EXISTS (SELECT 1 FROM links l0 WHERE l0.src = l.src AND l0.target = l.target AND l0.embed = 0)`;\n\nexport function toEdges(rows: Array<{ src: string; dst: string }>): [string, string][] {\n  return rows.map((r) => [r.src, r.dst]);\n}\n\n// Resolved edges, for rank (inside reconcile's afterReconcile pass).\nexport async function linkEdges(db: Connection): Promise<[string, string][]> {\n  const stmt = await db.prepare(LINK_EDGES_SQL);\n  return toEdges((await stmt.all()) as Array<{ src: string; dst: string }>);\n}\n\n// remove()/store() have already deleted rows by the time afterReconcile runs, so this records\n// whether any reconcile pass deleted at least one link row -- whether a file vanished, or a\n// reparse's stale rows shrank. dstChanged alone misses the reparsed-to-zero-links case (no\n// surviving rows to re-resolve). Keyed on the delta object, so the state dies with the reconcile.\nconst removedWithLinks = new WeakMap<ReconcileDelta, boolean>();\n\nexport const links: Feature = {\n  name: 'links',\n  async schema(db) {\n    await db.exec('CREATE TABLE IF NOT EXISTS links (src TEXT, target TEXT, target_base TEXT, dst TEXT, embed INTEGER, PRIMARY KEY (src, target, embed))');\n    await db.exec('CREATE INDEX IF NOT EXISTS links_dst ON links(dst)');\n    await db.exec('CREATE INDEX IF NOT EXISTS links_target_base ON links(target_base)');\n  },\n  extract,\n  // Reparsed docs are diffed in store() below instead of wiped here: deleting and re-inserting\n  // resets dst to NULL, which made every touch-only reparse read as an edge change and\n  // recompute PageRank -- measured as most of the remaining update cost. Only genuinely\n  // vanished files clear here.\n  async remove(db, paths, delta) {\n    const vanishedSet = new Set(delta.vanished);\n    const vanished = paths.filter((p) => vanishedSet.has(p));\n    if (vanished.length === 0) return;\n    const existingRows = await selectIn(db, 'src', vanished);\n    await db.runBatch(\n      'DELETE FROM links WHERE src = ?',\n      vanished.map((p) => [p])\n    );\n    if (existingRows.length > 0) removedWithLinks.set(delta, true);\n  },\n  async store(db, docs: ExtractedDoc[], delta) {\n    const byPath = new Map(docs.map((d) => [d.path, d.extracted as Array<{ target: string; embed: boolean }>]));\n    const addedSet = new Set(delta.added);\n    const reparsedExisting = docs.map((d) => d.path).filter((p) => !addedSet.has(p));\n\n    // Stale rows (existing target/embed pairs the new parse no longer contains) are real edge\n    // removals; diffed in JS from one bulk SELECT rather than a per-file NOT IN statement, so\n    // this whole pass costs one query plus one batched DELETE regardless of file count.\n    const existingRows = await selectIn(db, 'src', reparsedExisting);\n    const staleRows = existingRows.filter((row) => {\n      const targets = byPath.get(row.src) ?? [];\n      return !targets.some((t) => t.target === row.target && (t.embed ? 1 : 0) === row.embed);\n    });\n    if (staleRows.length > 0) {\n      await db.runBatch(\n        'DELETE FROM links WHERE src = ? AND target = ? AND embed = ?',\n        staleRows.map((r) => [r.src, r.target, r.embed])\n      );\n      removedWithLinks.set(delta, true);\n    }\n\n    // Upsert preserves dst on surviving rows, so an unchanged link resolves to the same value\n    // and reports no change; only genuinely new rows start at NULL. A same-note anchor's dst is\n    // always its own src, never another file's basename, so it gets no target_base -- SQL's\n    // `IN` never matches NULL, keeping it out of resolveIncremental's target_base lookup.\n    const upsertRows: unknown[][] = [];\n    for (const [path, targets] of byPath) {\n      for (const { target, embed } of targets) upsertRows.push([path, target, target.startsWith('#') ? null : baseKey(cleanTarget(target)), embed ? 1 : 0]);\n    }\n    if (upsertRows.length > 0) {\n      await db.runBatch('INSERT INTO links (src, target, target_base, dst, embed) VALUES (?, ?, ?, NULL, ?) ON CONFLICT(src, target, embed) DO UPDATE SET target_base = excluded.target_base', upsertRows);\n    }\n  },\n  async afterReconcile(db, delta) {\n    // Any deleted rows mean the edge set shrank -- whether the file vanished or was reparsed\n    // down to fewer/no links.\n    const removeHadLinks = removedWithLinks.get(delta) ?? false;\n\n    const churn = delta.reparsed.length + delta.vanished.length;\n    // Past this share of the tree, a full pass is the only one guaranteed to match\n    // resolveTarget's ambiguity rules. A cold build clears the threshold on its own.\n    const large = delta.files.length === 0 || churn > 0.2 * delta.files.length;\n\n    const dstChanged = large ? await resolveAll(db, delta.files) : await resolveIncremental(db, delta);\n    delta.linksChanged = dstChanged || removeHadLinks;\n  },\n};\n"],"names":["posix","maskRegions","parseWikilinkInner","inner","beforeAlias","split","trim","startsWith","length","target","extract","_raw","rawBody","_search","data","body","test","seen","Set","results","add","embed","key","has","push","m","matchAll","index","dest","includes","value","Object","values","item","Array","isArray","exec","cleanTarget","replace","baseKey","path","basename","toLowerCase","resolveTarget","src","pathSet","byBase","clean","fromSrc","normalize","join","dirname","candidate","candidates","get","buildByBase","files","Map","paths","map","f","relPath","sort","a","b","list","set","resolveRows","rows","changed","row","dst","writeResolved","db","runBatch","r","resolveAll","stmt","prepare","all","selectIn","column","keys","i","chunk","slice","resolveIncremental","delta","changedBasenames","p","added","vanished","collect","reparsed","LINK_EDGES_SQL","toEdges","linkEdges","removedWithLinks","WeakMap","links","name","schema","remove","vanishedSet","filter","existingRows","store","docs","byPath","d","extracted","addedSet","reparsedExisting","staleRows","targets","some","t","upsertRows","afterReconcile","removeHadLinks","churn","large","dstChanged","linksChanged"],"mappings":"AAAA,OAAOA,WAAW,kBAAkB;AAGpC,SAASC,WAAW,QAAQ,cAAc;AAG1C,0FAA0F;AAC1F,iFAAiF;AACjF,+FAA+F;AAC/F,uFAAuF;AAEvF,wFAAwF;AACxF,6FAA6F;AAC7F,uDAAuD;AACvD,uFAAuF;AACvF,yFAAyF;AACzF,4CAA4C;AAC5C,SAASC,mBAAmBC,KAAa;IACvC,MAAMC,cAAcD,MAAME,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;IAC5C,IAAIF,YAAYG,UAAU,CAAC,MAAM,OAAOH,YAAYI,MAAM,GAAG,IAAIJ,cAAc;IAC/E,MAAMK,SAASL,YAAYC,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;IAC7C,OAAOG,UAAU;AACnB;AAEA,SAASC,QAAQC,IAAY,EAAEC,OAAe,EAAEC,OAA4C,EAAEC,IAA8B;IAC1H,MAAMC,OAAO,YAAYC,IAAI,CAACJ,WAAWX,YAAYW,WAAWA;IAChE,MAAMK,OAAO,IAAIC;IACjB,MAAMC,UAAqD,EAAE;IAC7D,MAAMC,MAAM,CAACX,QAAgBY;QAC3B,MAAMC,MAAM,GAAGb,OAAO,EAAE,EAAEY,QAAQ,MAAM,KAAK;QAC7C,IAAIJ,KAAKM,GAAG,CAACD,MAAM;QACnBL,KAAKG,GAAG,CAACE;QACTH,QAAQK,IAAI,CAAC;YAAEf;YAAQY;QAAM;IAC/B;IACA,uFAAuF;IACvF,6DAA6D;IAC7D,KAAK,MAAMI,KAAKV,KAAKW,QAAQ,CAAC,kBAAmB;YAEjBD;QAD9B,MAAMhB,SAASP,mBAAmBuB,CAAC,CAAC,EAAE;QACtC,IAAIhB,QAAQW,IAAIX,QAAQM,IAAI,CAAC,EAACU,WAAAA,EAAEE,KAAK,cAAPF,sBAAAA,WAAW,KAAK,EAAE,KAAK;IACvD;IACA,sFAAsF;IACtF,mFAAmF;IACnF,wFAAwF;IACxF,2DAA2D;IAC3D,0FAA0F;IAC1F,mEAAmE;IACnE,KAAK,MAAMA,KAAKV,KAAKW,QAAQ,CAAC,6EAA8E;QAC1G,MAAME,OAAOH,CAAC,CAAC,EAAE,CAACnB,IAAI;QACtB,qFAAqF;QACrF,oFAAoF;QACpF,IAAI,CAACsB,QAAQA,KAAKC,QAAQ,CAAC,QAAQ;QACnC,IAAI,oCAAoCb,IAAI,CAACY,OAAO;QACpD,MAAMnB,SAASmB,KAAKrB,UAAU,CAAC,OAAOqB,OAAOA,KAAKvB,KAAK,CAAC,IAAI,CAAC,EAAE;QAC/D,IAAII,QAAQW,IAAIX,QAAQgB,CAAC,CAAC,EAAE,KAAK;IACnC;IACA,KAAK,MAAMK,SAASC,OAAOC,MAAM,CAAClB,iBAAAA,kBAAAA,OAAQ,CAAC,GAAI;QAC7C,KAAK,MAAMmB,QAAQC,MAAMC,OAAO,CAACL,SAASA,QAAQ;YAACA;SAAM,CAAE;YACzD,IAAI,OAAOG,SAAS,UAAU;YAC9B,MAAMR,IAAI,iBAAiBW,IAAI,CAACH,KAAK3B,IAAI;YACzC,IAAI,CAACmB,GAAG;YACR,MAAMhB,SAASP,mBAAmBuB,CAAC,CAAC,EAAE;YACtC,IAAIhB,QAAQW,IAAIX,QAAQ;QAC1B;IACF;IACA,OAAOU;AACT;AAEA,SAASkB,YAAY5B,MAAc;IACjC,OAAOA,OAAO6B,OAAO,CAAC,OAAO,KAAKA,OAAO,CAAC,SAAS;AACrD;AAEA,SAASC,QAAQC,IAAY;IAC3B,OAAOxC,MAAMyC,QAAQ,CAACD,MAAMF,OAAO,CAAC,UAAU,IAAII,WAAW;AAC/D;AAEA,uFAAuF;AACvF,0FAA0F;AAC1F,kGAAkG;AAClG,SAASC,cAAcC,GAAW,EAAEnC,MAAc,EAAEoC,OAAoB,EAAEC,MAA6B;IACrG,4FAA4F;IAC5F,IAAIrC,OAAOF,UAAU,CAAC,MAAM,OAAOqC;IACnC,MAAMG,QAAQV,YAAY5B;IAC1B,MAAMuC,UAAUhD,MAAMiD,SAAS,CAACjD,MAAMkD,IAAI,CAAClD,MAAMmD,OAAO,CAACP,MAAMG;IAC/D,KAAK,MAAMK,aAAa;QAACL;QAAO,GAAGA,MAAM,GAAG,CAAC;QAAEC;QAAS,GAAGA,QAAQ,GAAG,CAAC;KAAC,CAAE;QACxE,IAAIH,QAAQtB,GAAG,CAAC6B,YAAY,OAAOA;IACrC;IACA,MAAMC,aAAaP,OAAOQ,GAAG,CAACf,QAAQQ;IACtC,IAAI,CAACM,YAAY,OAAO;IACxB,0FAA0F;IAC1F,+DAA+D;IAC/D,OAAOA,WAAWxB,QAAQ,CAACe,OAAOA,MAAMS,UAAU,CAAC,EAAE;AACvD;AAEA,6FAA6F;AAC7F,yFAAyF;AACzF,SAASE,YAAYC,KAAiB;IACpC,MAAMV,SAAS,IAAIW;IACnB,MAAMC,QAAQF,MAAMG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO,EAAEC,IAAI,CAAC,CAACC,GAAGC,IAAMD,EAAEvD,MAAM,GAAGwD,EAAExD,MAAM,IAAKuD,CAAAA,IAAIC,IAAI,CAAC,IAAID,IAAIC,IAAI,IAAI,CAAA;IAC1G,KAAK,MAAMxB,QAAQkB,MAAO;QACxB,MAAMpC,MAAMiB,QAAQC;QACpB,MAAMyB,OAAOnB,OAAOQ,GAAG,CAAChC;QACxB,IAAI2C,MAAMA,KAAKzC,IAAI,CAACgB;aACfM,OAAOoB,GAAG,CAAC5C,KAAK;YAACkB;SAAK;IAC7B;IACA,OAAOM;AACT;AASA,6FAA6F;AAC7F,2FAA2F;AAC3F,6CAA6C;AAC7C,SAASqB,YAAYC,IAAe,EAAEvB,OAAoB,EAAEC,MAA6B;IACvF,MAAMuB,UAAqB,EAAE;IAC7B,KAAK,MAAMC,OAAOF,KAAM;QACtB,MAAMG,MAAM5B,cAAc2B,IAAI1B,GAAG,EAAE0B,IAAI7D,MAAM,EAAEoC,SAASC;QACxD,IAAIyB,QAAQD,IAAIC,GAAG,EAAEF,QAAQ7C,IAAI,CAAC;YAAE,GAAG8C,GAAG;YAAEC;QAAI;IAClD;IACA,OAAOF;AACT;AAEA,eAAeG,cAAcC,EAAc,EAAEL,IAAe;IAC1D,IAAIA,KAAK5D,MAAM,KAAK,GAAG;IACvB,MAAMiE,GAAGC,QAAQ,CACf,yDACAN,KAAKT,GAAG,CAAC,CAACgB,IAAM;YAACA,EAAEJ,GAAG;YAAEI,EAAE/B,GAAG;YAAE+B,EAAElE,MAAM;SAAC;AAE5C;AAEA,yFAAyF;AACzF,+EAA+E;AAC/E,eAAemE,WAAWH,EAAc,EAAEjB,KAAiB;IACzD,MAAMX,UAAU,IAAI3B,IAAIsC,MAAMG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IAClD,MAAMf,SAASS,YAAYC;IAC3B,MAAMqB,OAAO,MAAMJ,GAAGK,OAAO,CAAC;IAC9B,MAAMV,OAAQ,MAAMS,KAAKE,GAAG;IAC5B,MAAMV,UAAUF,YAAYC,MAAMvB,SAASC;IAC3C,MAAM0B,cAAcC,IAAIJ;IACxB,OAAOA,QAAQ7D,MAAM,GAAG;AAC1B;AAEA,2FAA2F;AAC3F,2FAA2F;AAC3F,eAAewE,SAASP,EAAc,EAAEQ,MAAc,EAAEC,IAAc;IACpE,MAAMd,OAAkB,EAAE;IAC1B,IAAK,IAAIe,IAAI,GAAGA,IAAID,KAAK1E,MAAM,EAAE2E,KAAK,IAAK;QACzC,MAAMC,QAAQF,KAAKG,KAAK,CAACF,GAAGA,IAAI;QAChC,MAAMN,OAAO,MAAMJ,GAAGK,OAAO,CAAC,CAAC,gDAAgD,EAAEG,OAAO,KAAK,EAAEG,MAAMzB,GAAG,CAAC,IAAM,KAAKT,IAAI,CAAC,MAAM,CAAC,CAAC;QACjIkB,KAAK5C,IAAI,IAAM,MAAMqD,KAAKE,GAAG,IAAIK;IACnC;IACA,OAAOhB;AACT;AAEA,yFAAyF;AACzF,yFAAyF;AACzF,kEAAkE;AAClE,eAAekB,mBAAmBb,EAAc,EAAEc,KAAqB;IACrE,MAAM1C,UAAU,IAAI3B,IAAIqE,MAAM/B,KAAK,CAACG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IACxD,MAAMf,SAASS,YAAYgC,MAAM/B,KAAK;IAEtC,MAAMgC,mBAAmB,IAAItE;IAC7B,KAAK,MAAMuE,KAAKF,MAAMG,KAAK,CAAEF,iBAAiBpE,GAAG,CAACmB,QAAQkD;IAC1D,KAAK,MAAMA,KAAKF,MAAMI,QAAQ,CAAEH,iBAAiBpE,GAAG,CAACmB,QAAQkD;IAE7D,+EAA+E;IAC/E,MAAMpC,aAAa,IAAII;IACvB,MAAMmC,UAAU,CAACxB;QACf,KAAK,MAAME,OAAOF,KAAMf,WAAWa,GAAG,CAAC,GAAGI,IAAI1B,GAAG,CAAC,EAAE,EAAE0B,IAAI7D,MAAM,CAAC,EAAE,EAAE6D,IAAIjD,KAAK,EAAE,EAAEiD;IACpF;IACAsB,QAAQ,MAAMZ,SAASP,IAAI,OAAOc,MAAMM,QAAQ;IAChDD,QAAQ,MAAMZ,SAASP,IAAI,eAAe;WAAIe;KAAiB;IAC/DI,QAAQ,MAAMZ,SAASP,IAAI,OAAOc,MAAMI,QAAQ;IAEhD,MAAMtB,UAAUF,YAAY;WAAId,WAAWrB,MAAM;KAAG,EAAEa,SAASC;IAC/D,MAAM0B,cAAcC,IAAIJ;IACxB,OAAOA,QAAQ7D,MAAM,GAAG;AAC1B;AAEA,0FAA0F;AAC1F,yFAAyF;AACzF,yFAAyF;AACzF,0FAA0F;AAC1F,iFAAiF;AACjF,4FAA4F;AAC5F,4FAA4F;AAC5F,6FAA6F;AAC7F,gFAAgF;AAChF,OAAO,MAAMsF,iBAAiB,CAAC;;;4GAG6E,CAAC,CAAC;AAE9G,OAAO,SAASC,QAAQ3B,IAAyC;IAC/D,OAAOA,KAAKT,GAAG,CAAC,CAACgB,IAAM;YAACA,EAAE/B,GAAG;YAAE+B,EAAEJ,GAAG;SAAC;AACvC;AAEA,qEAAqE;AACrE,OAAO,eAAeyB,UAAUvB,EAAc;IAC5C,MAAMI,OAAO,MAAMJ,GAAGK,OAAO,CAACgB;IAC9B,OAAOC,QAAS,MAAMlB,KAAKE,GAAG;AAChC;AAEA,8FAA8F;AAC9F,4FAA4F;AAC5F,2FAA2F;AAC3F,kGAAkG;AAClG,MAAMkB,mBAAmB,IAAIC;AAE7B,OAAO,MAAMC,QAAiB;IAC5BC,MAAM;IACN,MAAMC,QAAO5B,EAAE;QACb,MAAMA,GAAGrC,IAAI,CAAC;QACd,MAAMqC,GAAGrC,IAAI,CAAC;QACd,MAAMqC,GAAGrC,IAAI,CAAC;IAChB;IACA1B;IACA,6FAA6F;IAC7F,qFAAqF;IACrF,sFAAsF;IACtF,6BAA6B;IAC7B,MAAM4F,QAAO7B,EAAE,EAAEf,KAAK,EAAE6B,KAAK;QAC3B,MAAMgB,cAAc,IAAIrF,IAAIqE,MAAMI,QAAQ;QAC1C,MAAMA,WAAWjC,MAAM8C,MAAM,CAAC,CAACf,IAAMc,YAAYhF,GAAG,CAACkE;QACrD,IAAIE,SAASnF,MAAM,KAAK,GAAG;QAC3B,MAAMiG,eAAe,MAAMzB,SAASP,IAAI,OAAOkB;QAC/C,MAAMlB,GAAGC,QAAQ,CACf,mCACAiB,SAAShC,GAAG,CAAC,CAAC8B,IAAM;gBAACA;aAAE;QAEzB,IAAIgB,aAAajG,MAAM,GAAG,GAAGyF,iBAAiB/B,GAAG,CAACqB,OAAO;IAC3D;IACA,MAAMmB,OAAMjC,EAAE,EAAEkC,IAAoB,EAAEpB,KAAK;QACzC,MAAMqB,SAAS,IAAInD,IAAIkD,KAAKhD,GAAG,CAAC,CAACkD,IAAM;gBAACA,EAAErE,IAAI;gBAAEqE,EAAEC,SAAS;aAA8C;QACzG,MAAMC,WAAW,IAAI7F,IAAIqE,MAAMG,KAAK;QACpC,MAAMsB,mBAAmBL,KAAKhD,GAAG,CAAC,CAACkD,IAAMA,EAAErE,IAAI,EAAEgE,MAAM,CAAC,CAACf,IAAM,CAACsB,SAASxF,GAAG,CAACkE;QAE7E,0FAA0F;QAC1F,0FAA0F;QAC1F,oFAAoF;QACpF,MAAMgB,eAAe,MAAMzB,SAASP,IAAI,OAAOuC;QAC/C,MAAMC,YAAYR,aAAaD,MAAM,CAAC,CAAClC;gBACrBsC;YAAhB,MAAMM,WAAUN,cAAAA,OAAOtD,GAAG,CAACgB,IAAI1B,GAAG,eAAlBgE,yBAAAA,cAAuB,EAAE;YACzC,OAAO,CAACM,QAAQC,IAAI,CAAC,CAACC,IAAMA,EAAE3G,MAAM,KAAK6D,IAAI7D,MAAM,IAAI,AAAC2G,CAAAA,EAAE/F,KAAK,GAAG,IAAI,CAAA,MAAOiD,IAAIjD,KAAK;QACxF;QACA,IAAI4F,UAAUzG,MAAM,GAAG,GAAG;YACxB,MAAMiE,GAAGC,QAAQ,CACf,gEACAuC,UAAUtD,GAAG,CAAC,CAACgB,IAAM;oBAACA,EAAE/B,GAAG;oBAAE+B,EAAElE,MAAM;oBAAEkE,EAAEtD,KAAK;iBAAC;YAEjD4E,iBAAiB/B,GAAG,CAACqB,OAAO;QAC9B;QAEA,0FAA0F;QAC1F,4FAA4F;QAC5F,wFAAwF;QACxF,sFAAsF;QACtF,MAAM8B,aAA0B,EAAE;QAClC,KAAK,MAAM,CAAC7E,MAAM0E,QAAQ,IAAIN,OAAQ;YACpC,KAAK,MAAM,EAAEnG,MAAM,EAAEY,KAAK,EAAE,IAAI6F,QAASG,WAAW7F,IAAI,CAAC;gBAACgB;gBAAM/B;gBAAQA,OAAOF,UAAU,CAAC,OAAO,OAAOgC,QAAQF,YAAY5B;gBAAUY,QAAQ,IAAI;aAAE;QACtJ;QACA,IAAIgG,WAAW7G,MAAM,GAAG,GAAG;YACzB,MAAMiE,GAAGC,QAAQ,CAAC,uKAAuK2C;QAC3L;IACF;IACA,MAAMC,gBAAe7C,EAAE,EAAEc,KAAK;YAGLU;QAFvB,yFAAyF;QACzF,0BAA0B;QAC1B,MAAMsB,kBAAiBtB,wBAAAA,iBAAiB3C,GAAG,CAACiC,oBAArBU,mCAAAA,wBAA+B;QAEtD,MAAMuB,QAAQjC,MAAMM,QAAQ,CAACrF,MAAM,GAAG+E,MAAMI,QAAQ,CAACnF,MAAM;QAC3D,+EAA+E;QAC/E,iFAAiF;QACjF,MAAMiH,QAAQlC,MAAM/B,KAAK,CAAChD,MAAM,KAAK,KAAKgH,QAAQ,MAAMjC,MAAM/B,KAAK,CAAChD,MAAM;QAE1E,MAAMkH,aAAaD,QAAQ,MAAM7C,WAAWH,IAAIc,MAAM/B,KAAK,IAAI,MAAM8B,mBAAmBb,IAAIc;QAC5FA,MAAMoC,YAAY,GAAGD,cAAcH;IACrC;AACF,EAAE"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/links.ts"],"sourcesContent":["import posix from 'node:path/posix';\nimport type { FileStat } from '../scan/index.ts';\nimport type { Connection } from '../store/types.ts';\nimport { maskRegions } from './fences.ts';\nimport type { ExtractedDoc, Feature, ReconcileDelta } from './types.ts';\n\n// links(src, target, target_base, dst, embed): target as written, target_base its indexed baseKey, dst the resolved path or NULL (a queryable dead link).\n// embed distinguishes `![[x]]`/`![](x.md)` from `[[x]]`/`[](x.md)` -- Obsidian's grain: a target both linked and embedded in one note is two rows.\n\n// Alias stripped after |; a leading # keeps the anchor as the target (self-link needs a\n// heading name), else the anchor splits off. Null = not a link.\nfunction parseWikilinkInner(inner: string): string | null {\n  const beforeAlias = inner.split('|')[0].trim();\n  if (beforeAlias.startsWith('#')) return beforeAlias.length > 1 ? beforeAlias : null;\n  const target = beforeAlias.split('#')[0].trim();\n  return target || null;\n}\n\n// A frontmatter value counts only when it is exactly a wikilink (\"[[X]]\") -- mid-string and\n// ![[...]] forms don't count there, per Obsidian's rule (probe-verified).\nfunction extract(_raw: string, rawBody: string, _search?: { title: string; summary: string }, data?: Record<string, unknown>): Array<{ target: string; embed: boolean }> {\n  const body = /\\[\\[|\\]\\(/.test(rawBody) ? maskRegions(rawBody) : rawBody;\n  const seen = new Set<string>();\n  const results: Array<{ target: string; embed: boolean }> = [];\n  const add = (target: string, embed: boolean) => {\n    const key = `${target}\\0${embed ? '1' : '0'}`;\n    if (seen.has(key)) return;\n    seen.add(key);\n    results.push({ target, embed });\n  };\n  // To the first ]], as Obsidian parses it, so a heading or alias holding a lone ] still\n  // matches; anchor and alias split off in parseWikilinkInner.\n  for (const m of body.matchAll(/\\[\\[(.*?)\\]\\]/g)) {\n    const target = parseWikilinkInner(m[1]);\n    if (target) add(target, body[(m.index ?? 0) - 1] === '!');\n  }\n  // Any internal destination, not only .md-suffixed: \\](Zektor) resolves like [[Zektor]] and\n  // \\](#anchor) is a self-edge. External URLs and titled links (dest stops at whitespace) skip.\n  for (const m of body.matchAll(/(!)?\\[[^\\]]*\\]\\(((?:[^()\\s]|\\([^()\\s]*\\))+)(?:\\s+(?:\"[^)]*\"|'[^)]*'))?\\)/g)) {\n    const dest = m[2].trim();\n    // External: a URI scheme prefix (mailto:, tel:, data:, https:...), protocol-relative\n    // //host, a www. shorthand, or a scheme anywhere (the malformed double-paren case).\n    if (!dest || dest.includes('://')) continue;\n    if (/^([a-z][a-z0-9+.-]*:|\\/\\/|www\\.)/i.test(dest)) continue;\n    const target = dest.startsWith('#') ? dest : dest.split('#')[0];\n    if (target) add(target, m[1] === '!');\n  }\n  for (const value of Object.values(data ?? {})) {\n    for (const item of Array.isArray(value) ? value : [value]) {\n      if (typeof item !== 'string') continue;\n      const m = /^\\[\\[(.*)\\]\\]$/.exec(item.trim());\n      if (!m) continue;\n      const target = parseWikilinkInner(m[1]);\n      if (target) add(target, false);\n    }\n  }\n  return results;\n}\n\nfunction cleanTarget(target: string): string {\n  return target.replace(/\\\\/g, '/').replace(/^\\.\\//, '');\n}\n\nfunction baseKey(path: string): string {\n  return posix.basename(path).replace(/\\.md$/i, '').toLowerCase();\n}\n\n// Obsidian-style: exact relative path (with/without .md), path relative to the linking note's\n// directory, then basename match (shortest path wins ties, verified against a cold-loaded vault).\nfunction resolveTarget(src: string, target: string, pathSet: Set<string>, byBase: Map<string, string[]>): string | null {\n  // [[#Heading]]: a same-note anchor, always the note itself -- never depends on other files.\n  if (target.startsWith('#')) return src;\n  const clean = cleanTarget(target);\n  const fromSrc = posix.normalize(posix.join(posix.dirname(src), clean));\n  for (const candidate of [clean, `${clean}.md`, fromSrc, `${fromSrc}.md`]) {\n    if (pathSet.has(candidate)) return candidate;\n  }\n  const candidates = byBase.get(baseKey(clean));\n  if (!candidates) return null;\n  // The linking note itself wins a basename collision (Obsidian resolves to self before the\n  // shortest-path rule -- cold-load verified on the hub corpus).\n  return candidates.includes(src) ? src : candidates[0];\n}\n\n// Shortest path wins a basename collision; equal lengths fall back to lexicographic, our own\n// deterministic tiebreak for a case Obsidian itself leaves registration-order-dependent.\nfunction buildByBase(files: FileStat[]): Map<string, string[]> {\n  const byBase = new Map<string, string[]>();\n  const paths = files.map((f) => f.relPath).sort((a, b) => a.length - b.length || (a < b ? -1 : a > b ? 1 : 0));\n  for (const path of paths) {\n    const key = baseKey(path);\n    const list = byBase.get(key);\n    if (list) list.push(path);\n    else byBase.set(key, [path]);\n  }\n  return byBase;\n}\n\ninterface LinkRow {\n  src: string;\n  target: string;\n  dst: string | null;\n  embed: number;\n}\n\n// resolveTarget applied to `rows`, returning only the rows whose dst changed, so the caller\n// can write them in one runBatch call.\nfunction resolveRows(rows: LinkRow[], pathSet: Set<string>, byBase: Map<string, string[]>): LinkRow[] {\n  const changed: LinkRow[] = [];\n  for (const row of rows) {\n    const dst = resolveTarget(row.src, row.target, pathSet, byBase);\n    if (dst !== row.dst) changed.push({ ...row, dst });\n  }\n  return changed;\n}\n\nasync function writeResolved(db: Connection, rows: LinkRow[]): Promise<void> {\n  if (rows.length === 0) return;\n  await db.runBatch(\n    'UPDATE links SET dst = ? WHERE src = ? AND target = ?',\n    rows.map((r) => [r.dst, r.src, r.target])\n  );\n}\n\n// A new or deleted file can change any note's resolution, so re-resolve the whole table.\n// Fallback for cold builds and large deltas -- see afterReconcile's threshold.\nasync function resolveAll(db: Connection, files: FileStat[]): Promise<boolean> {\n  const pathSet = new Set(files.map((f) => f.relPath));\n  const byBase = buildByBase(files);\n  const stmt = await db.prepare('SELECT src, target, dst, embed FROM links');\n  const rows = (await stmt.all()) as unknown as LinkRow[];\n  const changed = resolveRows(rows, pathSet, byBase);\n  await writeResolved(db, changed);\n  return changed.length > 0;\n}\n\n// Chunked so a large-but-under-threshold delta never exceeds SQLite's bound-variable limit\n// (SQLITE_MAX_VARIABLE_NUMBER, 32766); harmless headroom on DuckDB, which has no such cap.\nasync function selectIn(db: Connection, column: string, keys: string[]): Promise<LinkRow[]> {\n  const rows: LinkRow[] = [];\n  for (let i = 0; i < keys.length; i += 500) {\n    const chunk = keys.slice(i, i + 500);\n    const stmt = await db.prepare(`SELECT src, target, dst, embed FROM links WHERE ${column} IN (${chunk.map(() => '?').join(', ')})`);\n    rows.push(...((await stmt.all(...chunk)) as unknown as LinkRow[]));\n  }\n  return rows;\n}\n\n// Re-resolves only what this reconcile could have touched: re-stored sources, links whose\n// target basename matches an added/vanished file, and links whose dst vanished -- all indexed.\nasync function resolveIncremental(db: Connection, delta: ReconcileDelta): Promise<boolean> {\n  const pathSet = new Set(delta.files.map((f) => f.relPath));\n  const byBase = buildByBase(delta.files);\n\n  const changedBasenames = new Set<string>();\n  for (const p of delta.added) changedBasenames.add(baseKey(p));\n  for (const p of delta.vanished) changedBasenames.add(baseKey(p));\n\n  // \\0 can't appear in a path or target, so the key never collides; a space can.\n  const candidates = new Map<string, LinkRow>();\n  const collect = (rows: LinkRow[]) => {\n    for (const row of rows) candidates.set(`${row.src}\\0${row.target}\\0${row.embed}`, row);\n  };\n  collect(await selectIn(db, 'src', delta.reparsed));\n  collect(await selectIn(db, 'target_base', [...changedBasenames]));\n  collect(await selectIn(db, 'dst', delta.vanished));\n\n  const changed = resolveRows([...candidates.values()], pathSet, byBase);\n  await writeResolved(db, changed);\n  return changed.length > 0;\n}\n\n// One edge per (src, target); an embed row is excluded only when the same pair also exists as\n// a link. Self-edges are dropped so PageRank mass is not self-recycled.\nexport const LINK_EDGES_SQL = `SELECT src, dst FROM links WHERE dst IS NOT NULL AND embed = 0 AND src != dst\n    UNION ALL\n    SELECT src, dst FROM links l WHERE dst IS NOT NULL AND embed = 1 AND src != dst\n      AND NOT EXISTS (SELECT 1 FROM links l0 WHERE l0.src = l.src AND l0.target = l.target AND l0.embed = 0)`;\n\nexport function toEdges(rows: Array<{ src: string; dst: string }>): [string, string][] {\n  return rows.map((r) => [r.src, r.dst]);\n}\n\n// Resolved edges, for rank (inside reconcile's afterReconcile pass).\nexport async function linkEdges(db: Connection): Promise<[string, string][]> {\n  const stmt = await db.prepare(LINK_EDGES_SQL);\n  return toEdges((await stmt.all()) as Array<{ src: string; dst: string }>);\n}\n\n// Records whether remove()/store() deleted any link row, since afterReconcile runs after both\n// and dstChanged alone misses the reparsed-to-zero-links case. Keyed on delta so state dies with it.\nconst removedWithLinks = new WeakMap<ReconcileDelta, boolean>();\n\nexport const links: Feature = {\n  name: 'links',\n  async schema(db) {\n    await db.exec('CREATE TABLE IF NOT EXISTS links (src TEXT, target TEXT, target_base TEXT, dst TEXT, embed INTEGER, PRIMARY KEY (src, target, embed))');\n    await db.exec('CREATE INDEX IF NOT EXISTS links_dst ON links(dst)');\n    await db.exec('CREATE INDEX IF NOT EXISTS links_target_base ON links(target_base)');\n  },\n  extract,\n  // Reparsed docs are diffed in store() instead of wiped here: delete-and-reinsert resets dst\n  // to NULL, making every touch-only reparse look like an edge change and recompute PageRank.\n  async remove(db, paths, delta) {\n    const vanishedSet = new Set(delta.vanished);\n    const vanished = paths.filter((p) => vanishedSet.has(p));\n    if (vanished.length === 0) return;\n    const existingRows = await selectIn(db, 'src', vanished);\n    await db.runBatch(\n      'DELETE FROM links WHERE src = ?',\n      vanished.map((p) => [p])\n    );\n    if (existingRows.length > 0) removedWithLinks.set(delta, true);\n  },\n  async store(db, docs: ExtractedDoc[], delta) {\n    const byPath = new Map(docs.map((d) => [d.path, d.extracted as Array<{ target: string; embed: boolean }>]));\n    const addedSet = new Set(delta.added);\n    const reparsedExisting = docs.map((d) => d.path).filter((p) => !addedSet.has(p));\n\n    // Stale rows (target/embed pairs missing from the new parse) are diffed in JS from one\n    // bulk SELECT, so this pass costs one query plus one batched DELETE regardless of file count.\n    const existingRows = await selectIn(db, 'src', reparsedExisting);\n    const staleRows = existingRows.filter((row) => {\n      const targets = byPath.get(row.src) ?? [];\n      return !targets.some((t) => t.target === row.target && (t.embed ? 1 : 0) === row.embed);\n    });\n    if (staleRows.length > 0) {\n      await db.runBatch(\n        'DELETE FROM links WHERE src = ? AND target = ? AND embed = ?',\n        staleRows.map((r) => [r.src, r.target, r.embed])\n      );\n      removedWithLinks.set(delta, true);\n    }\n\n    // Upsert preserves dst on surviving rows; only new rows start at NULL. A same-note anchor\n    // gets no target_base, so SQL's `IN` excludes it from resolveIncremental's lookup.\n    const upsertRows: unknown[][] = [];\n    for (const [path, targets] of byPath) {\n      for (const { target, embed } of targets) upsertRows.push([path, target, target.startsWith('#') ? null : baseKey(cleanTarget(target)), embed ? 1 : 0]);\n    }\n    if (upsertRows.length > 0) {\n      await db.runBatch('INSERT INTO links (src, target, target_base, dst, embed) VALUES (?, ?, ?, NULL, ?) ON CONFLICT(src, target, embed) DO UPDATE SET target_base = excluded.target_base', upsertRows);\n    }\n  },\n  async afterReconcile(db, delta) {\n    // Any deleted rows mean the edge set shrank -- whether the file vanished or was reparsed\n    // down to fewer/no links.\n    const removeHadLinks = removedWithLinks.get(delta) ?? false;\n\n    const churn = delta.reparsed.length + delta.vanished.length;\n    // Past this share of the tree, a full pass is the only one guaranteed to match\n    // resolveTarget's ambiguity rules. A cold build clears the threshold on its own.\n    const large = delta.files.length === 0 || churn > 0.2 * delta.files.length;\n\n    const dstChanged = large ? await resolveAll(db, delta.files) : await resolveIncremental(db, delta);\n    delta.linksChanged = dstChanged || removeHadLinks;\n  },\n};\n"],"names":["posix","maskRegions","parseWikilinkInner","inner","beforeAlias","split","trim","startsWith","length","target","extract","_raw","rawBody","_search","data","body","test","seen","Set","results","add","embed","key","has","push","m","matchAll","index","dest","includes","value","Object","values","item","Array","isArray","exec","cleanTarget","replace","baseKey","path","basename","toLowerCase","resolveTarget","src","pathSet","byBase","clean","fromSrc","normalize","join","dirname","candidate","candidates","get","buildByBase","files","Map","paths","map","f","relPath","sort","a","b","list","set","resolveRows","rows","changed","row","dst","writeResolved","db","runBatch","r","resolveAll","stmt","prepare","all","selectIn","column","keys","i","chunk","slice","resolveIncremental","delta","changedBasenames","p","added","vanished","collect","reparsed","LINK_EDGES_SQL","toEdges","linkEdges","removedWithLinks","WeakMap","links","name","schema","remove","vanishedSet","filter","existingRows","store","docs","byPath","d","extracted","addedSet","reparsedExisting","staleRows","targets","some","t","upsertRows","afterReconcile","removeHadLinks","churn","large","dstChanged","linksChanged"],"mappings":"AAAA,OAAOA,WAAW,kBAAkB;AAGpC,SAASC,WAAW,QAAQ,cAAc;AAG1C,0JAA0J;AAC1J,mJAAmJ;AAEnJ,wFAAwF;AACxF,gEAAgE;AAChE,SAASC,mBAAmBC,KAAa;IACvC,MAAMC,cAAcD,MAAME,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;IAC5C,IAAIF,YAAYG,UAAU,CAAC,MAAM,OAAOH,YAAYI,MAAM,GAAG,IAAIJ,cAAc;IAC/E,MAAMK,SAASL,YAAYC,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;IAC7C,OAAOG,UAAU;AACnB;AAEA,4FAA4F;AAC5F,0EAA0E;AAC1E,SAASC,QAAQC,IAAY,EAAEC,OAAe,EAAEC,OAA4C,EAAEC,IAA8B;IAC1H,MAAMC,OAAO,YAAYC,IAAI,CAACJ,WAAWX,YAAYW,WAAWA;IAChE,MAAMK,OAAO,IAAIC;IACjB,MAAMC,UAAqD,EAAE;IAC7D,MAAMC,MAAM,CAACX,QAAgBY;QAC3B,MAAMC,MAAM,GAAGb,OAAO,EAAE,EAAEY,QAAQ,MAAM,KAAK;QAC7C,IAAIJ,KAAKM,GAAG,CAACD,MAAM;QACnBL,KAAKG,GAAG,CAACE;QACTH,QAAQK,IAAI,CAAC;YAAEf;YAAQY;QAAM;IAC/B;IACA,uFAAuF;IACvF,6DAA6D;IAC7D,KAAK,MAAMI,KAAKV,KAAKW,QAAQ,CAAC,kBAAmB;YAEjBD;QAD9B,MAAMhB,SAASP,mBAAmBuB,CAAC,CAAC,EAAE;QACtC,IAAIhB,QAAQW,IAAIX,QAAQM,IAAI,CAAC,EAACU,WAAAA,EAAEE,KAAK,cAAPF,sBAAAA,WAAW,KAAK,EAAE,KAAK;IACvD;IACA,2FAA2F;IAC3F,8FAA8F;IAC9F,KAAK,MAAMA,KAAKV,KAAKW,QAAQ,CAAC,6EAA8E;QAC1G,MAAME,OAAOH,CAAC,CAAC,EAAE,CAACnB,IAAI;QACtB,qFAAqF;QACrF,oFAAoF;QACpF,IAAI,CAACsB,QAAQA,KAAKC,QAAQ,CAAC,QAAQ;QACnC,IAAI,oCAAoCb,IAAI,CAACY,OAAO;QACpD,MAAMnB,SAASmB,KAAKrB,UAAU,CAAC,OAAOqB,OAAOA,KAAKvB,KAAK,CAAC,IAAI,CAAC,EAAE;QAC/D,IAAII,QAAQW,IAAIX,QAAQgB,CAAC,CAAC,EAAE,KAAK;IACnC;IACA,KAAK,MAAMK,SAASC,OAAOC,MAAM,CAAClB,iBAAAA,kBAAAA,OAAQ,CAAC,GAAI;QAC7C,KAAK,MAAMmB,QAAQC,MAAMC,OAAO,CAACL,SAASA,QAAQ;YAACA;SAAM,CAAE;YACzD,IAAI,OAAOG,SAAS,UAAU;YAC9B,MAAMR,IAAI,iBAAiBW,IAAI,CAACH,KAAK3B,IAAI;YACzC,IAAI,CAACmB,GAAG;YACR,MAAMhB,SAASP,mBAAmBuB,CAAC,CAAC,EAAE;YACtC,IAAIhB,QAAQW,IAAIX,QAAQ;QAC1B;IACF;IACA,OAAOU;AACT;AAEA,SAASkB,YAAY5B,MAAc;IACjC,OAAOA,OAAO6B,OAAO,CAAC,OAAO,KAAKA,OAAO,CAAC,SAAS;AACrD;AAEA,SAASC,QAAQC,IAAY;IAC3B,OAAOxC,MAAMyC,QAAQ,CAACD,MAAMF,OAAO,CAAC,UAAU,IAAII,WAAW;AAC/D;AAEA,8FAA8F;AAC9F,kGAAkG;AAClG,SAASC,cAAcC,GAAW,EAAEnC,MAAc,EAAEoC,OAAoB,EAAEC,MAA6B;IACrG,4FAA4F;IAC5F,IAAIrC,OAAOF,UAAU,CAAC,MAAM,OAAOqC;IACnC,MAAMG,QAAQV,YAAY5B;IAC1B,MAAMuC,UAAUhD,MAAMiD,SAAS,CAACjD,MAAMkD,IAAI,CAAClD,MAAMmD,OAAO,CAACP,MAAMG;IAC/D,KAAK,MAAMK,aAAa;QAACL;QAAO,GAAGA,MAAM,GAAG,CAAC;QAAEC;QAAS,GAAGA,QAAQ,GAAG,CAAC;KAAC,CAAE;QACxE,IAAIH,QAAQtB,GAAG,CAAC6B,YAAY,OAAOA;IACrC;IACA,MAAMC,aAAaP,OAAOQ,GAAG,CAACf,QAAQQ;IACtC,IAAI,CAACM,YAAY,OAAO;IACxB,0FAA0F;IAC1F,+DAA+D;IAC/D,OAAOA,WAAWxB,QAAQ,CAACe,OAAOA,MAAMS,UAAU,CAAC,EAAE;AACvD;AAEA,6FAA6F;AAC7F,yFAAyF;AACzF,SAASE,YAAYC,KAAiB;IACpC,MAAMV,SAAS,IAAIW;IACnB,MAAMC,QAAQF,MAAMG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO,EAAEC,IAAI,CAAC,CAACC,GAAGC,IAAMD,EAAEvD,MAAM,GAAGwD,EAAExD,MAAM,IAAKuD,CAAAA,IAAIC,IAAI,CAAC,IAAID,IAAIC,IAAI,IAAI,CAAA;IAC1G,KAAK,MAAMxB,QAAQkB,MAAO;QACxB,MAAMpC,MAAMiB,QAAQC;QACpB,MAAMyB,OAAOnB,OAAOQ,GAAG,CAAChC;QACxB,IAAI2C,MAAMA,KAAKzC,IAAI,CAACgB;aACfM,OAAOoB,GAAG,CAAC5C,KAAK;YAACkB;SAAK;IAC7B;IACA,OAAOM;AACT;AASA,4FAA4F;AAC5F,uCAAuC;AACvC,SAASqB,YAAYC,IAAe,EAAEvB,OAAoB,EAAEC,MAA6B;IACvF,MAAMuB,UAAqB,EAAE;IAC7B,KAAK,MAAMC,OAAOF,KAAM;QACtB,MAAMG,MAAM5B,cAAc2B,IAAI1B,GAAG,EAAE0B,IAAI7D,MAAM,EAAEoC,SAASC;QACxD,IAAIyB,QAAQD,IAAIC,GAAG,EAAEF,QAAQ7C,IAAI,CAAC;YAAE,GAAG8C,GAAG;YAAEC;QAAI;IAClD;IACA,OAAOF;AACT;AAEA,eAAeG,cAAcC,EAAc,EAAEL,IAAe;IAC1D,IAAIA,KAAK5D,MAAM,KAAK,GAAG;IACvB,MAAMiE,GAAGC,QAAQ,CACf,yDACAN,KAAKT,GAAG,CAAC,CAACgB,IAAM;YAACA,EAAEJ,GAAG;YAAEI,EAAE/B,GAAG;YAAE+B,EAAElE,MAAM;SAAC;AAE5C;AAEA,yFAAyF;AACzF,+EAA+E;AAC/E,eAAemE,WAAWH,EAAc,EAAEjB,KAAiB;IACzD,MAAMX,UAAU,IAAI3B,IAAIsC,MAAMG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IAClD,MAAMf,SAASS,YAAYC;IAC3B,MAAMqB,OAAO,MAAMJ,GAAGK,OAAO,CAAC;IAC9B,MAAMV,OAAQ,MAAMS,KAAKE,GAAG;IAC5B,MAAMV,UAAUF,YAAYC,MAAMvB,SAASC;IAC3C,MAAM0B,cAAcC,IAAIJ;IACxB,OAAOA,QAAQ7D,MAAM,GAAG;AAC1B;AAEA,2FAA2F;AAC3F,2FAA2F;AAC3F,eAAewE,SAASP,EAAc,EAAEQ,MAAc,EAAEC,IAAc;IACpE,MAAMd,OAAkB,EAAE;IAC1B,IAAK,IAAIe,IAAI,GAAGA,IAAID,KAAK1E,MAAM,EAAE2E,KAAK,IAAK;QACzC,MAAMC,QAAQF,KAAKG,KAAK,CAACF,GAAGA,IAAI;QAChC,MAAMN,OAAO,MAAMJ,GAAGK,OAAO,CAAC,CAAC,gDAAgD,EAAEG,OAAO,KAAK,EAAEG,MAAMzB,GAAG,CAAC,IAAM,KAAKT,IAAI,CAAC,MAAM,CAAC,CAAC;QACjIkB,KAAK5C,IAAI,IAAM,MAAMqD,KAAKE,GAAG,IAAIK;IACnC;IACA,OAAOhB;AACT;AAEA,0FAA0F;AAC1F,+FAA+F;AAC/F,eAAekB,mBAAmBb,EAAc,EAAEc,KAAqB;IACrE,MAAM1C,UAAU,IAAI3B,IAAIqE,MAAM/B,KAAK,CAACG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IACxD,MAAMf,SAASS,YAAYgC,MAAM/B,KAAK;IAEtC,MAAMgC,mBAAmB,IAAItE;IAC7B,KAAK,MAAMuE,KAAKF,MAAMG,KAAK,CAAEF,iBAAiBpE,GAAG,CAACmB,QAAQkD;IAC1D,KAAK,MAAMA,KAAKF,MAAMI,QAAQ,CAAEH,iBAAiBpE,GAAG,CAACmB,QAAQkD;IAE7D,+EAA+E;IAC/E,MAAMpC,aAAa,IAAII;IACvB,MAAMmC,UAAU,CAACxB;QACf,KAAK,MAAME,OAAOF,KAAMf,WAAWa,GAAG,CAAC,GAAGI,IAAI1B,GAAG,CAAC,EAAE,EAAE0B,IAAI7D,MAAM,CAAC,EAAE,EAAE6D,IAAIjD,KAAK,EAAE,EAAEiD;IACpF;IACAsB,QAAQ,MAAMZ,SAASP,IAAI,OAAOc,MAAMM,QAAQ;IAChDD,QAAQ,MAAMZ,SAASP,IAAI,eAAe;WAAIe;KAAiB;IAC/DI,QAAQ,MAAMZ,SAASP,IAAI,OAAOc,MAAMI,QAAQ;IAEhD,MAAMtB,UAAUF,YAAY;WAAId,WAAWrB,MAAM;KAAG,EAAEa,SAASC;IAC/D,MAAM0B,cAAcC,IAAIJ;IACxB,OAAOA,QAAQ7D,MAAM,GAAG;AAC1B;AAEA,8FAA8F;AAC9F,wEAAwE;AACxE,OAAO,MAAMsF,iBAAiB,CAAC;;;4GAG6E,CAAC,CAAC;AAE9G,OAAO,SAASC,QAAQ3B,IAAyC;IAC/D,OAAOA,KAAKT,GAAG,CAAC,CAACgB,IAAM;YAACA,EAAE/B,GAAG;YAAE+B,EAAEJ,GAAG;SAAC;AACvC;AAEA,qEAAqE;AACrE,OAAO,eAAeyB,UAAUvB,EAAc;IAC5C,MAAMI,OAAO,MAAMJ,GAAGK,OAAO,CAACgB;IAC9B,OAAOC,QAAS,MAAMlB,KAAKE,GAAG;AAChC;AAEA,8FAA8F;AAC9F,qGAAqG;AACrG,MAAMkB,mBAAmB,IAAIC;AAE7B,OAAO,MAAMC,QAAiB;IAC5BC,MAAM;IACN,MAAMC,QAAO5B,EAAE;QACb,MAAMA,GAAGrC,IAAI,CAAC;QACd,MAAMqC,GAAGrC,IAAI,CAAC;QACd,MAAMqC,GAAGrC,IAAI,CAAC;IAChB;IACA1B;IACA,4FAA4F;IAC5F,4FAA4F;IAC5F,MAAM4F,QAAO7B,EAAE,EAAEf,KAAK,EAAE6B,KAAK;QAC3B,MAAMgB,cAAc,IAAIrF,IAAIqE,MAAMI,QAAQ;QAC1C,MAAMA,WAAWjC,MAAM8C,MAAM,CAAC,CAACf,IAAMc,YAAYhF,GAAG,CAACkE;QACrD,IAAIE,SAASnF,MAAM,KAAK,GAAG;QAC3B,MAAMiG,eAAe,MAAMzB,SAASP,IAAI,OAAOkB;QAC/C,MAAMlB,GAAGC,QAAQ,CACf,mCACAiB,SAAShC,GAAG,CAAC,CAAC8B,IAAM;gBAACA;aAAE;QAEzB,IAAIgB,aAAajG,MAAM,GAAG,GAAGyF,iBAAiB/B,GAAG,CAACqB,OAAO;IAC3D;IACA,MAAMmB,OAAMjC,EAAE,EAAEkC,IAAoB,EAAEpB,KAAK;QACzC,MAAMqB,SAAS,IAAInD,IAAIkD,KAAKhD,GAAG,CAAC,CAACkD,IAAM;gBAACA,EAAErE,IAAI;gBAAEqE,EAAEC,SAAS;aAA8C;QACzG,MAAMC,WAAW,IAAI7F,IAAIqE,MAAMG,KAAK;QACpC,MAAMsB,mBAAmBL,KAAKhD,GAAG,CAAC,CAACkD,IAAMA,EAAErE,IAAI,EAAEgE,MAAM,CAAC,CAACf,IAAM,CAACsB,SAASxF,GAAG,CAACkE;QAE7E,uFAAuF;QACvF,8FAA8F;QAC9F,MAAMgB,eAAe,MAAMzB,SAASP,IAAI,OAAOuC;QAC/C,MAAMC,YAAYR,aAAaD,MAAM,CAAC,CAAClC;gBACrBsC;YAAhB,MAAMM,WAAUN,cAAAA,OAAOtD,GAAG,CAACgB,IAAI1B,GAAG,eAAlBgE,yBAAAA,cAAuB,EAAE;YACzC,OAAO,CAACM,QAAQC,IAAI,CAAC,CAACC,IAAMA,EAAE3G,MAAM,KAAK6D,IAAI7D,MAAM,IAAI,AAAC2G,CAAAA,EAAE/F,KAAK,GAAG,IAAI,CAAA,MAAOiD,IAAIjD,KAAK;QACxF;QACA,IAAI4F,UAAUzG,MAAM,GAAG,GAAG;YACxB,MAAMiE,GAAGC,QAAQ,CACf,gEACAuC,UAAUtD,GAAG,CAAC,CAACgB,IAAM;oBAACA,EAAE/B,GAAG;oBAAE+B,EAAElE,MAAM;oBAAEkE,EAAEtD,KAAK;iBAAC;YAEjD4E,iBAAiB/B,GAAG,CAACqB,OAAO;QAC9B;QAEA,0FAA0F;QAC1F,mFAAmF;QACnF,MAAM8B,aAA0B,EAAE;QAClC,KAAK,MAAM,CAAC7E,MAAM0E,QAAQ,IAAIN,OAAQ;YACpC,KAAK,MAAM,EAAEnG,MAAM,EAAEY,KAAK,EAAE,IAAI6F,QAASG,WAAW7F,IAAI,CAAC;gBAACgB;gBAAM/B;gBAAQA,OAAOF,UAAU,CAAC,OAAO,OAAOgC,QAAQF,YAAY5B;gBAAUY,QAAQ,IAAI;aAAE;QACtJ;QACA,IAAIgG,WAAW7G,MAAM,GAAG,GAAG;YACzB,MAAMiE,GAAGC,QAAQ,CAAC,uKAAuK2C;QAC3L;IACF;IACA,MAAMC,gBAAe7C,EAAE,EAAEc,KAAK;YAGLU;QAFvB,yFAAyF;QACzF,0BAA0B;QAC1B,MAAMsB,kBAAiBtB,wBAAAA,iBAAiB3C,GAAG,CAACiC,oBAArBU,mCAAAA,wBAA+B;QAEtD,MAAMuB,QAAQjC,MAAMM,QAAQ,CAACrF,MAAM,GAAG+E,MAAMI,QAAQ,CAACnF,MAAM;QAC3D,+EAA+E;QAC/E,iFAAiF;QACjF,MAAMiH,QAAQlC,MAAM/B,KAAK,CAAChD,MAAM,KAAK,KAAKgH,QAAQ,MAAMjC,MAAM/B,KAAK,CAAChD,MAAM;QAE1E,MAAMkH,aAAaD,QAAQ,MAAM7C,WAAWH,IAAIc,MAAM/B,KAAK,IAAI,MAAM8B,mBAAmBb,IAAIc;QAC5FA,MAAMoC,YAAY,GAAGD,cAAcH;IACrC;AACF,EAAE"}

package/dist/esm/features/sections.js

@@ -1,25 +1,23 @@
-import { estimateTokens } from '../chunk/index.js';
-import { fenceTracker } from './fences.js';
-// Headings outside fenced code blocks.
-function extract(raw) {
-    const lines = raw.split('\n');
-    const found = [];
-    const fence = fenceTracker();
-    for(let i = 0; i < lines.length; i++){
-        if (fence.feed(lines[i])) continue;
-        if (fence.inFence) continue;
-        const m = lines[i].match(/^(#{1,6}) +(.*)/);
-        if (m) found.push({
-            level: m[1].length,
-            heading: m[2].trim(),
-            startLine: i + 1,
-            endLine: lines.length,
+import { estimateTokens, parse } from '../chunk/index.js';
+import { countLines } from '../scan/frontmatter.js';
+// Heading blocks parse() already found (mdast/CommonMark fences), offset back onto the raw file;
+// a section runs to just before the next heading, or EOF.
+function sectionsFromBlocks(blocks, raw, body) {
+    const rawLines = raw.split('\n');
+    const offset = rawLines.length - countLines(body);
+    const found = blocks.filter((b)=>b.type === 'heading').map((b)=>{
+        var _b_depth, _b_text;
+        return {
+            level: (_b_depth = b.depth) !== null && _b_depth !== void 0 ? _b_depth : 1,
+            heading: ((_b_text = b.text) !== null && _b_text !== void 0 ? _b_text : '').trim(),
+            startLine: b.startLine + offset,
+            endLine: rawLines.length,
             tokens: 0
-        });
-    }
+        };
+    });
     for(let s = 0; s < found.length; s++){
         if (s + 1 < found.length) found[s].endLine = found[s + 1].startLine - 1;
-        const text = lines.slice(found[s].startLine - 1, found[s].endLine).join('\n');
+        const text = rawLines.slice(found[s].startLine - 1, found[s].endLine).join('\n');
         found[s].tokens = Math.ceil(estimateTokens(text));
     }
     return found;
@@ -29,7 +27,9 @@ export const sections = {
     async schema (db) {
         await db.exec(`CREATE TABLE IF NOT EXISTS sections ("path" TEXT, idx INTEGER, level INTEGER, heading TEXT, start_line INTEGER, end_line INTEGER, tokens INTEGER, PRIMARY KEY ("path", idx))`);
     },
-    extract,
+    extract (raw, body, _search, _data, _cfg, blocks) {
+        return sectionsFromBlocks(blocks !== null && blocks !== void 0 ? blocks : parse(body), raw, body);
+    },
     async remove (db, paths) {
         if (paths.length === 0) return;
         await db.runBatch('DELETE FROM sections WHERE "path" = ?', paths.map((p)=>[

package/dist/esm/features/sections.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/sections.ts"],"sourcesContent":["import { estimateTokens } from '../chunk/index.ts';\nimport { fenceTracker } from './fences.ts';\nimport type { Feature } from './types.ts';\n\n// sections(path, idx, level, heading, start_line, end_line, tokens): the heading outline,\n// 1-indexed over the raw file so a row is a direct Read range; tokens is estimateTokens (D5),\n// CJK-aware rather than a flat chars/4.\n\nexport interface Section {\n  level: number;\n  heading: string;\n  startLine: number;\n  endLine: number;\n  tokens: number;\n}\n\n// Headings outside fenced code blocks.\nfunction extract(raw: string): Section[] {\n  const lines = raw.split('\\n');\n  const found: Section[] = [];\n  const fence = fenceTracker();\n  for (let i = 0; i < lines.length; i++) {\n    if (fence.feed(lines[i])) continue;\n    if (fence.inFence) continue;\n    const m = lines[i].match(/^(#{1,6}) +(.*)/);\n    if (m) found.push({ level: m[1].length, heading: m[2].trim(), startLine: i + 1, endLine: lines.length, tokens: 0 });\n  }\n  for (let s = 0; s < found.length; s++) {\n    if (s + 1 < found.length) found[s].endLine = found[s + 1].startLine - 1;\n    const text = lines.slice(found[s].startLine - 1, found[s].endLine).join('\\n');\n    found[s].tokens = Math.ceil(estimateTokens(text));\n  }\n  return found;\n}\n\nexport const sections: Feature = {\n  name: 'sections',\n  async schema(db) {\n    await db.exec(`CREATE TABLE IF NOT EXISTS sections (\"path\" TEXT, idx INTEGER, level INTEGER, heading TEXT, start_line INTEGER, end_line INTEGER, tokens INTEGER, PRIMARY KEY (\"path\", idx))`);\n  },\n  extract,\n  async remove(db, paths) {\n    if (paths.length === 0) return;\n    await db.runBatch(\n      'DELETE FROM sections WHERE \"path\" = ?',\n      paths.map((p) => [p])\n    );\n  },\n  async store(db, docs) {\n    const rows: unknown[][] = [];\n    for (const { path, extracted } of docs) (extracted as Section[]).forEach((s, idx) => rows.push([path, idx, s.level, s.heading, s.startLine, s.endLine, s.tokens]));\n    if (rows.length === 0) return;\n    await db.runBatch('INSERT INTO sections (\"path\", idx, level, heading, start_line, end_line, tokens) VALUES (?, ?, ?, ?, ?, ?, ?)', rows);\n  },\n};\n"],"names":["estimateTokens","fenceTracker","extract","raw","lines","split","found","fence","i","length","feed","inFence","m","match","push","level","heading","trim","startLine","endLine","tokens","s","text","slice","join","Math","ceil","sections","name","schema","db","exec","remove","paths","runBatch","map","p","store","docs","rows","path","extracted","forEach","idx"],"mappings":"AAAA,SAASA,cAAc,QAAQ,oBAAoB;AACnD,SAASC,YAAY,QAAQ,cAAc;AAe3C,uCAAuC;AACvC,SAASC,QAAQC,GAAW;IAC1B,MAAMC,QAAQD,IAAIE,KAAK,CAAC;IACxB,MAAMC,QAAmB,EAAE;IAC3B,MAAMC,QAAQN;IACd,IAAK,IAAIO,IAAI,GAAGA,IAAIJ,MAAMK,MAAM,EAAED,IAAK;QACrC,IAAID,MAAMG,IAAI,CAACN,KAAK,CAACI,EAAE,GAAG;QAC1B,IAAID,MAAMI,OAAO,EAAE;QACnB,MAAMC,IAAIR,KAAK,CAACI,EAAE,CAACK,KAAK,CAAC;QACzB,IAAID,GAAGN,MAAMQ,IAAI,CAAC;YAAEC,OAAOH,CAAC,CAAC,EAAE,CAACH,MAAM;YAAEO,SAASJ,CAAC,CAAC,EAAE,CAACK,IAAI;YAAIC,WAAWV,IAAI;YAAGW,SAASf,MAAMK,MAAM;YAAEW,QAAQ;QAAE;IACnH;IACA,IAAK,IAAIC,IAAI,GAAGA,IAAIf,MAAMG,MAAM,EAAEY,IAAK;QACrC,IAAIA,IAAI,IAAIf,MAAMG,MAAM,EAAEH,KAAK,CAACe,EAAE,CAACF,OAAO,GAAGb,KAAK,CAACe,IAAI,EAAE,CAACH,SAAS,GAAG;QACtE,MAAMI,OAAOlB,MAAMmB,KAAK,CAACjB,KAAK,CAACe,EAAE,CAACH,SAAS,GAAG,GAAGZ,KAAK,CAACe,EAAE,CAACF,OAAO,EAAEK,IAAI,CAAC;QACxElB,KAAK,CAACe,EAAE,CAACD,MAAM,GAAGK,KAAKC,IAAI,CAAC1B,eAAesB;IAC7C;IACA,OAAOhB;AACT;AAEA,OAAO,MAAMqB,WAAoB;IAC/BC,MAAM;IACN,MAAMC,QAAOC,EAAE;QACb,MAAMA,GAAGC,IAAI,CAAC,CAAC,4KAA4K,CAAC;IAC9L;IACA7B;IACA,MAAM8B,QAAOF,EAAE,EAAEG,KAAK;QACpB,IAAIA,MAAMxB,MAAM,KAAK,GAAG;QACxB,MAAMqB,GAAGI,QAAQ,CACf,yCACAD,MAAME,GAAG,CAAC,CAACC,IAAM;gBAACA;aAAE;IAExB;IACA,MAAMC,OAAMP,EAAE,EAAEQ,IAAI;QAClB,MAAMC,OAAoB,EAAE;QAC5B,KAAK,MAAM,EAAEC,IAAI,EAAEC,SAAS,EAAE,IAAIH,KAAM,AAACG,UAAwBC,OAAO,CAAC,CAACrB,GAAGsB,MAAQJ,KAAKzB,IAAI,CAAC;gBAAC0B;gBAAMG;gBAAKtB,EAAEN,KAAK;gBAAEM,EAAEL,OAAO;gBAAEK,EAAEH,SAAS;gBAAEG,EAAEF,OAAO;gBAAEE,EAAED,MAAM;aAAC;QAChK,IAAImB,KAAK9B,MAAM,KAAK,GAAG;QACvB,MAAMqB,GAAGI,QAAQ,CAAC,iHAAiHK;IACrI;AACF,EAAE"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/sections.ts"],"sourcesContent":["import type { Block } from '../chunk/index.ts';\nimport { estimateTokens, parse } from '../chunk/index.ts';\nimport { countLines } from '../scan/frontmatter.ts';\nimport type { Feature } from './types.ts';\n\n// sections(path, idx, level, heading, start_line, end_line, tokens): the heading outline,\n// 1-indexed over the raw file so a row is a direct Read range; tokens is estimateTokens (D5), CJK-aware.\n\nexport interface Section {\n  level: number;\n  heading: string;\n  startLine: number;\n  endLine: number;\n  tokens: number;\n}\n\n// Heading blocks parse() already found (mdast/CommonMark fences), offset back onto the raw file;\n// a section runs to just before the next heading, or EOF.\nfunction sectionsFromBlocks(blocks: Block[], raw: string, body: string): Section[] {\n  const rawLines = raw.split('\\n');\n  const offset = rawLines.length - countLines(body);\n  const found: Section[] = blocks.filter((b) => b.type === 'heading').map((b) => ({ level: b.depth ?? 1, heading: (b.text ?? '').trim(), startLine: b.startLine + offset, endLine: rawLines.length, tokens: 0 }));\n  for (let s = 0; s < found.length; s++) {\n    if (s + 1 < found.length) found[s].endLine = found[s + 1].startLine - 1;\n    const text = rawLines.slice(found[s].startLine - 1, found[s].endLine).join('\\n');\n    found[s].tokens = Math.ceil(estimateTokens(text));\n  }\n  return found;\n}\n\nexport const sections: Feature = {\n  name: 'sections',\n  async schema(db) {\n    await db.exec(`CREATE TABLE IF NOT EXISTS sections (\"path\" TEXT, idx INTEGER, level INTEGER, heading TEXT, start_line INTEGER, end_line INTEGER, tokens INTEGER, PRIMARY KEY (\"path\", idx))`);\n  },\n  extract(raw, body, _search, _data, _cfg, blocks) {\n    return sectionsFromBlocks(blocks ?? parse(body), raw, body);\n  },\n  async remove(db, paths) {\n    if (paths.length === 0) return;\n    await db.runBatch(\n      'DELETE FROM sections WHERE \"path\" = ?',\n      paths.map((p) => [p])\n    );\n  },\n  async store(db, docs) {\n    const rows: unknown[][] = [];\n    for (const { path, extracted } of docs) (extracted as Section[]).forEach((s, idx) => rows.push([path, idx, s.level, s.heading, s.startLine, s.endLine, s.tokens]));\n    if (rows.length === 0) return;\n    await db.runBatch('INSERT INTO sections (\"path\", idx, level, heading, start_line, end_line, tokens) VALUES (?, ?, ?, ?, ?, ?, ?)', rows);\n  },\n};\n"],"names":["estimateTokens","parse","countLines","sectionsFromBlocks","blocks","raw","body","rawLines","split","offset","length","found","filter","b","type","map","level","depth","heading","text","trim","startLine","endLine","tokens","s","slice","join","Math","ceil","sections","name","schema","db","exec","extract","_search","_data","_cfg","remove","paths","runBatch","p","store","docs","rows","path","extracted","forEach","idx","push"],"mappings":"AACA,SAASA,cAAc,EAAEC,KAAK,QAAQ,oBAAoB;AAC1D,SAASC,UAAU,QAAQ,yBAAyB;AAcpD,iGAAiG;AACjG,0DAA0D;AAC1D,SAASC,mBAAmBC,MAAe,EAAEC,GAAW,EAAEC,IAAY;IACpE,MAAMC,WAAWF,IAAIG,KAAK,CAAC;IAC3B,MAAMC,SAASF,SAASG,MAAM,GAAGR,WAAWI;IAC5C,MAAMK,QAAmBP,OAAOQ,MAAM,CAAC,CAACC,IAAMA,EAAEC,IAAI,KAAK,WAAWC,GAAG,CAAC,CAACF;YAAgBA,UAAwBA;eAAjC;YAAEG,KAAK,GAAEH,WAAAA,EAAEI,KAAK,cAAPJ,sBAAAA,WAAW;YAAGK,SAAS,EAACL,UAAAA,EAAEM,IAAI,cAANN,qBAAAA,UAAU,IAAIO,IAAI;YAAIC,WAAWR,EAAEQ,SAAS,GAAGZ;YAAQa,SAASf,SAASG,MAAM;YAAEa,QAAQ;QAAE;;IAC5M,IAAK,IAAIC,IAAI,GAAGA,IAAIb,MAAMD,MAAM,EAAEc,IAAK;QACrC,IAAIA,IAAI,IAAIb,MAAMD,MAAM,EAAEC,KAAK,CAACa,EAAE,CAACF,OAAO,GAAGX,KAAK,CAACa,IAAI,EAAE,CAACH,SAAS,GAAG;QACtE,MAAMF,OAAOZ,SAASkB,KAAK,CAACd,KAAK,CAACa,EAAE,CAACH,SAAS,GAAG,GAAGV,KAAK,CAACa,EAAE,CAACF,OAAO,EAAEI,IAAI,CAAC;QAC3Ef,KAAK,CAACa,EAAE,CAACD,MAAM,GAAGI,KAAKC,IAAI,CAAC5B,eAAemB;IAC7C;IACA,OAAOR;AACT;AAEA,OAAO,MAAMkB,WAAoB;IAC/BC,MAAM;IACN,MAAMC,QAAOC,EAAE;QACb,MAAMA,GAAGC,IAAI,CAAC,CAAC,4KAA4K,CAAC;IAC9L;IACAC,SAAQ7B,GAAG,EAAEC,IAAI,EAAE6B,OAAO,EAAEC,KAAK,EAAEC,IAAI,EAAEjC,MAAM;QAC7C,OAAOD,mBAAmBC,mBAAAA,oBAAAA,SAAUH,MAAMK,OAAOD,KAAKC;IACxD;IACA,MAAMgC,QAAON,EAAE,EAAEO,KAAK;QACpB,IAAIA,MAAM7B,MAAM,KAAK,GAAG;QACxB,MAAMsB,GAAGQ,QAAQ,CACf,yCACAD,MAAMxB,GAAG,CAAC,CAAC0B,IAAM;gBAACA;aAAE;IAExB;IACA,MAAMC,OAAMV,EAAE,EAAEW,IAAI;QAClB,MAAMC,OAAoB,EAAE;QAC5B,KAAK,MAAM,EAAEC,IAAI,EAAEC,SAAS,EAAE,IAAIH,KAAM,AAACG,UAAwBC,OAAO,CAAC,CAACvB,GAAGwB,MAAQJ,KAAKK,IAAI,CAAC;gBAACJ;gBAAMG;gBAAKxB,EAAER,KAAK;gBAAEQ,EAAEN,OAAO;gBAAEM,EAAEH,SAAS;gBAAEG,EAAEF,OAAO;gBAAEE,EAAED,MAAM;aAAC;QAChK,IAAIqB,KAAKlC,MAAM,KAAK,GAAG;QACvB,MAAMsB,GAAGQ,QAAQ,CAAC,iHAAiHI;IACrI;AACF,EAAE"}

package/dist/esm/features/tags.js

@@ -1,7 +1,6 @@
 import { maskRegions } from './fences.js';
 // tags(path, tag): Obsidian's file.tags grain -- frontmatter list/string tags plus inline
-// #tags from the prose, deduplicated, source not distinguished. Nested tags store full
-// (book/scifi); `tag = 'book' OR tag LIKE 'book/%'` is how a caller matches the parent too.
+// #tags, deduplicated, source not distinguished. Nested tags store full (book/scifi); `tag = 'book' OR tag LIKE 'book/%'` matches the parent too.
 // Obsidian treats [[#Heading]] as a same-note link, not a tag.
 const WIKILINK_RE = /\[\[.*?\]\]/g; // to the first ]], so a heading holding a lone ] still masks
 // Obsidian doesn't read tags inside HTML markup.
@@ -32,9 +31,8 @@ function frontmatterTags(data) {
     }
     return found;
 }
-// #tag tokens outside fenced code blocks, inline code spans, wikilinks, HTML tags, HTML blocks,
-// <!-- --> comments, and link destinations. maskRegions() handles all of those region-level
-// concerns (and code spans, masked once there); this loop only does per-line tag extraction.
+// #tag tokens outside the regions maskRegions() masks (fences, code spans, wikilinks, HTML,
+// comments, link destinations); this loop only does per-line tag extraction.
 function inlineTags(body) {
     const found = [];
     for (const line of maskRegions(body).split('\n')){
@@ -65,9 +63,8 @@ export const tags = {
         await db.exec('CREATE INDEX IF NOT EXISTS tags_tag ON tags(tag)');
     },
     extract,
-    // Blanket clear-then-reinsert (store() below always writes the full found list back), so
-    // remove() covers both a vanished file and a reparsed one that already had rows -- no
-    // per-file NOT IN diffing, which cannot be expressed as one statement shared across files.
+    // Blanket clear-then-reinsert (store() always writes the full found list back), so remove()
+    // covers vanished and reparsed files without per-file NOT IN diffing, which cannot be one statement shared across files.
     async remove (db, paths) {
         if (paths.length === 0) return;
         await db.runBatch('DELETE FROM tags WHERE "path" = ?', paths.map((p)=>[